## Why

This is the continuation of this #3507 to enable `CompartmentSet` reports in coreneuron. They are nested on top of `Compartment` reports. In fact, conceptually, they are a filtered version since we report the values typically found in a `compartment` report but only for some specific points; specified in an additional file. The specs are described [here](https://sonata-extension.readthedocs.io/en/latest/sonata_report.html#fine-grained-compartment-report).

Neurodamus/neuron already handle most of this. However, coreneuron generates reports by itself (using libsonatareport) so we need to add support in coreneuron itself. 

## How

Qualitatively this is done in a few steps:

- add a way to pass the points to coreneuron because it does not read `compartment_sets.json` by itself. This is done by adding a few lines of binary values to `report.conf`. A point is defined as a triplet: (`gid`, `section_id`, `compartment_id`) where:
- `gid` is the usual gid of a cell
- `section_id` is the progressive id of the sections. Local of each cell and it depends on the stection types that are on the cell. Qualitatively, it uniquely defines a section on the cell and you can always pass for example `dend[16] <-> section_id`.
- `compartment_id` I am not 100% sure of how these ids are really taken. However I see that I can: 
    - pick them from a `seg` (compartment) with `soma[0](0.5).node_index()`
    - they are non-repeating per thread and can identify a `node`. A node is an element in one of the basic neuron matrices 
    - differently from the nodes, they are not contiguous. The mapping can be retrieveed from `Memb_list.nodeindices` where: `Memb_list.nodeindices[i_node] = compartment_id`. 
    - reversing `Memb_list.nodeindices` allows me to retrieve the node id to, finally, pick the correct `double*` necessary for libsoantareport to work. This lazy mapping is done in `fill_segment_id_2_node_id(`. `get_var(` is in charge of calling this when necessary. The only thing left to manage this cache is to decide when to clear the cache (every new gid gets a new cache). Fortunately gids are progressive so this is already optimal.

## Changelog

### `nrnreport.hpp`
- [x] add `CompartmentSet` report type
- [x] remove `TargetType` and split it in `Compartments` and `SectionType`: no need to encode/decode 2 enums into 1 in a non-critical part of the code
- [x] add a few fields to `ReportConfiguration` to manage `CompartmentSet` report and the removal of `TargetType`

### `report_configuration_parser.cpp`
- [x] simplify reading of `report.conf`
- [x] standardize the reading of binary `vector<int>` from `report.conf` with `fill_vec_int`
- [x] add a few `nrn_assert`: `compartment` reports cannot pick multiple variables for the same point

### `report_handler.cpp`
- [x] standardized errors and asserts
- [x] reorder function position to make most of the functions local in this translation unit: they were never used outside
- [x] lazy caching with `fill_segment_id_2_node_id`. Moved out of `get_var` to carify its purpose. In some particular cases it may be useful to have it as a stand-alone function
- [x] remove `check_section_type`: if the `ReportType` is unknown we throw an error directly in `create_report`
- [x] reworked `intersection_gids` to create a list of ids to pick in `report.target`. `CompartmentSet` can easily have repeated `gids` (not a thing before) which were discarded before. This made the previous list badly suited for a 3-vector id system as in `CompartmentSet`. It could have been done without this change but this was by far the simplest, cleanest, and most elegant solution. The change for the existing functions is minimal. From:
```
for (const auto& gid: gids_to_report) {
```
to:

```
 for (const auto& intersection_id: intersection_ids) {
        const auto gid = report.target[intersection_id];
 ```
- [x] remove `nodes_to_gids` and `map_gids` which is, now, obsolete and inefficient. It requires a loop over all the nodes instead of picking only the good ones.
- [x] `all_compartments` is not part of the configuration (in a nice enum: `Compartments`)

### `report_handler.hpp`
- [x] remove most of the member methods as they were used only in `report_handler.cpp`. Faster compilation. 

### `test/external/CMakeLists.txt`
- [x] updated external tests and its tag

## Context

This should finally take care of the compartment sets for coreneuron

Fix: #345 #322

testing for:

neuronsimulator/nrn#3507
neuronsimulator/nrn#3542

## Scope

- [x] rework report handling (mostly in node and target_manager this time. i.e. removal of `_report_setup`)
- [x] new `CoreReportConfig` to finally handle in a proper way `report.conf`
- [x] remove `write_sim_config`, `update_report_conf`, `write_report_conf` and similar functions in `CoreConfig`. Use `CoreSimulationConfig` and `CoreReportConfig` instead
- [x] move `cell-permute` in another issue/pr. EDIT done: #367
- [x] scaling moved to `report_parameters.py`
- [x] `CumulativeError` moved in `pyutils`
- [x] `ReportParams` now has its own file alleviating the burden of the `node.py` file. `_report_build_params` (previously in `node.py`) is also inside the new file.
- [x] `target_type` is no more. We do not need encoding/decoding. We just write `sections` and `compartments` in the `report.conf` 
- [x] streamlined `enable_reports`. It is still a complex function but it should be easier to read and with less edge cases 
- [x] use the decorator `@cache_errors` to just fill a `CumulativeError` instead of raising an error and raise it at the end
- [x] `merge_dicts` (in `tests/utils.py`) allows for overrides and suppression of sections from the child dict with the special keywords: `override_field` and `delete_field`

### `report.conf` changes

- [x] `type_name` has been split into `sections` and `compartments`. No need to encode everything in an enum to decode later. This simplifies code and makes the `report.conf`. The code is slightly slower `0(1)` in a part of the code that was not performance-critical
- [x] `scaling` added at the end of the report line (after buffer size)
- [x] add 2 additional binary lines in case of a compartment set report to pass the exact positions of the compartment sets. They are necessary if and only if the report type is `compartment_set`. Having them when it is not or not having them when it is is an error.

## Testing

- [x] add integration-e2e tests in `test_reports.py`

## Testing
- [x] test various bug fixes in coreneuron
- [x] add lfp test. Check that it fails with neuron
- [x] fix integration-e2e
- [x] fix scientific
- [x] fix unit

## Context

This should finally take care of the compartment sets for coreneuron

Fix: #345 #322

testing for:

neuronsimulator/nrn#3507
neuronsimulator/nrn#3542

## Scope

- [x] rework report handling (mostly in node and target_manager this time. i.e. removal of `_report_setup`)
- [x] new `CoreReportConfig` to finally handle in a proper way `report.conf`
- [x] remove `write_sim_config`, `update_report_conf`, `write_report_conf` and similar functions in `CoreConfig`. Use `CoreSimulationConfig` and `CoreReportConfig` instead
- [x] move `cell-permute` in another issue/pr. EDIT done: #367
- [x] scaling moved to `report_parameters.py`
- [x] `CumulativeError` moved in `pyutils`
- [x] `ReportParams` now has its own file alleviating the burden of the `node.py` file. `_report_build_params` (previously in `node.py`) is also inside the new file.
- [x] `target_type` is no more. We do not need encoding/decoding. We just write `sections` and `compartments` in the `report.conf` 
- [x] streamlined `enable_reports`. It is still a complex function but it should be easier to read and with less edge cases 
- [x] use the decorator `@cache_errors` to just fill a `CumulativeError` instead of raising an error and raise it at the end
- [x] `merge_dicts` (in `tests/utils.py`) allows for overrides and suppression of sections from the child dict with the special keywords: `override_field` and `delete_field`

### `report.conf` changes

- [x] `type_name` has been split into `sections` and `compartments`. No need to encode everything in an enum to decode later. This simplifies code and makes the `report.conf`. The code is slightly slower `0(1)` in a part of the code that was not performance-critical
- [x] `scaling` added at the end of the report line (after buffer size)
- [x] add 2 additional binary lines in case of a compartment set report to pass the exact positions of the compartment sets. They are necessary if and only if the report type is `compartment_set`. Having them when it is not or not having them when it is is an error.

## Testing

- [x] add integration-e2e tests in `test_reports.py`

## Testing
- [x] test various bug fixes in coreneuron
- [x] add lfp test. Check that it fails with neuron
- [x] fix integration-e2e
- [x] fix scientific
- [x] fix unit