…ons (#2029)

# What does this PR do?

Two improvements to the tracer FFI surface in `libdd-data-pipeline-ffi`:

1. **Check for nullable pointers.**  

2. **Return `ByteSlice` from `ddog_trace_exporter_response_get_body`.**
   Replace the separate `*const u8` return + `out_len` out-parameter with
   a single `ByteSlice` return value, matching the existing slice
   conventions in `libdd-common-ffi`.

# Motivation

FUP to #1952. C consumers (dd-trace-rb) can now use the existing
`read_ddogerr_string_and_drop` / `get_error_details_and_drop` helpers
from `datadog_ruby_common.h` instead of custom error handling for tracer
construction functions. The `ByteSlice` return avoids an awkward split
between return value and out-parameter for the response body.

# Additional Notes

AI was used to accelerate implementation; all code was reviewed and understood.

# How to test the change?

```
cargo test -p libdd-data-pipeline-ffi --lib
```

All 40 tests pass (13 tracer + 3 response + 24 trace_exporter).


Co-authored-by: hoolioh <julio.gonzalez@datadoghq.com>
Co-authored-by: ekump <edmund.kump@datadoghq.com>

#What does this PR do?

  Bumps the libdatadog workspace version from 34.0.0 → 35.0.0 (Cargo.toml + regenerated Cargo.lock) to cut the 35.0.0 release.

  Breaking changes affecting the FFI layer (34.0.0 → 35.0.0)

  1. Tracer FFI: response body now returned as ByteSlice (#2029) — feat!
  - ddog_trace_exporter_response_get_body no longer takes an out_len: *mut usize out-parameter and returns *const u8. It now returns a single ByteSlice value, matching the slice
  conventions in libdd-common-ffi.
  - Tracer constructor functions now validate nullable pointers and return ErrorCode::InvalidArgument instead of risking UB; panics across the boundary are caught and surfaced as
  ErrorCode::Panic.
  - C consumers (e.g. dd-trace-rb) can now reuse the standard read_ddogerr_string_and_drop / get_error_details_and_drop helpers from datadog_ruby_common.h.
  - Action for callers: drop the out_len argument and read length/pointer from the returned ByteSlice.

  2. Crashtracker: error.threads flattened to []ThreadData (#2054) — fix(crashtracking)!
  - The crash-report payload now emits error.threads as an optional flat array of thread objects (unified runtime stack schema 1.8) instead of the previous nested threads object.
  - This is a wire/payload-format change consumed by downstream pipelines and the product UI.
  - Action for callers: consumers parsing crash reports must update to the 1.8 thread layout.

  Other breaking changes since 34.0.0 (Rust API, not C ABI)

  These are !-marked breaking changes in libdd-trace-utils that affect Rust consumers of the library but do not change the C FFI surface or generated headers:

  - feat(trace-utils)!: introduce VecMap datastructure (#2022) — adds a HashMap replacement optimized for span construction (not yet wired into span fields).
  - feat(trace-utils)!: add from_string to span text (#2011) — requires From<String> on SpanText and switches SpanSlice to Cow<str> to allow span-field mutation/normalization.

  Notable new FFI surface (additive)

  - feat: Add FFI for trace exporter (#1952) — new tracer FFI surface in libdd-data-pipeline-ffi (the change above, #2029, refines it).


Co-authored-by: julio.gonzalez <julio.gonzalez@datadoghq.com>

…r trace exporter FFI (#2051)

# What does this PR do?

Add fork safety and cooperative cancellation support to the trace exporter FFI:

1. **Fork hooks**: Expose `ddog_trace_exporter_before_fork`, `_after_fork_in_parent`,
   and `_after_fork_in_child` that delegate to the underlying `SharedRuntime`. These
   allow C callers to coordinate the tokio runtime lifecycle around process forks.

2. **Cancellation token**: Introduce `ddog_trace_exporter_cancel_token_new`, `_cancel`,
   and `_drop` for managing cancellation tokens. Thread the token into
   `ddog_trace_exporter_send_trace_chunks` via `tokio::select!` so callers can
   cooperatively abort in-flight HTTP requests.

# Motivation

FUP to #1952.

Ruby application servers (Puma, Unicorn, Passenger) fork worker processes. The tokio
runtime does not survive `fork()` — its threads are not carried over. Without the fork
hooks, the trace exporter is dead in child processes.

The existing `RUBY_UBF_IO` unblock function on the dd-trace-rb side sends a signal
that cannot actually cancel an in-flight Rust HTTP request. The cancellation token
enables cooperative abort: when Ruby interrupts a thread (shutdown, `Thread#kill`),
the UBF cancels the token, which causes `tokio::select!` to abort the send and return
promptly.

# Additional Notes

- Added a `pub fn shared_runtime()` accessor to `TraceExporter` since the field was private.
- The cancellation token uses `Handle<CancellationToken>` from `libdd-common-ffi`, same
  pattern as profiling-ffi, with a `ddog_TraceExporterCancelToken` cbindgen rename to
  avoid symbol conflicts.
- The `cancel` parameter on `send_trace_chunks` is nullable; passing `NULL` preserves
  existing behavior.

AI was used to accelerate implementation; all code was reviewed and understood.

# How to test the change?

```
cargo test -p libdd-data-pipeline-ffi --lib
```

46 tests pass (40 existing + 6 new for fork hooks and cancellation token).

Co-authored-by: dd-octo-sts[bot] <200755185+dd-octo-sts[bot]@users.noreply.github.com>
Co-authored-by: VianneyRuhlmann <vianney.ruhlmann@datadoghq.com>