Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@integration-tests/tests/data/json_multifile/logs/sts-135-2011-07-11.jsonl`:
- Around line 1-8: The JSONL records for mission "STS-135" (each object with
"line_index" 0–7) currently have CRLF endings; change the file so every record
uses LF-only line endings (remove trailing \r from each line) ensuring no CR
characters remain at end-of-line and re-save the JSONL with Unix/LF EOL to
prevent byte-level comparison flakes.

In `@integration-tests/tests/data/text_multifile/logs/apollo-17_day10.txt`:
- Line 4: Replace the misspelled phrase "ground based" with the hyphenated form
"ground-based" in the fixture text line that reads "navigation sightings compare
well with ground based tracking solutions" so it becomes "navigation sightings
compare well with ground-based tracking solutions"; update the test fixture file
content accordingly to preserve surrounding punctuation and spacing.

In `@integration-tests/tests/fixtures/datasets.py`:
- Around line 17-23: The fixtures duplicate dataset-loading logic and access
nested metadata without validation, causing brittle failures; create a shared
loader/validator (e.g., a helper function like load_and_validate_dataset or
load_json_to_dataset) that centralizes path resolution using path_to_dataset and
load_json_to_dict, validates required keys such as
metadata_dict["file_structure"]["logs_subdir"] (and any schema fields), raises
clear errors on missing/invalid fields, and returns an IntegrationTestDataset
populated via dataset_name, path_to_dataset_logs and metadata_dict; replace the
duplicated blocks (including the other occurrence at lines 31-37) to call this
helper so failures are fast and diagnostics are meaningful.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@integration-tests/tests/data/text_multifile/metadata.json`:
- Line 16: The metadata file sets "columns_file_name": null but the
IntegrationTestDatasetMetadata model currently types columns_file_name as str;
update the model (IntegrationTestDatasetMetadata) to accept None by changing the
type to Optional[str] or using the union syntax (str | None) for
columns_file_name so Pydantic validation succeeds when the field is null.

In `@integration-tests/tests/utils/classes.py`:
- Line 75: The type annotation for columns_file_name must allow nulls; change
its declaration in the class to be nullable (use "str | None" or
"Optional[str]") so Pydantic accepts null from text_multifile/metadata.json;
also add or update the typing import (Optional) if you choose that form. Ensure
this aligns with the existing columns_file_path property which already handles
None.
- Line 96: The code calls validate_dir_exists on a file path
(metadata_file_path); replace this with a file-specific check: add or reuse a
validate_file_exists utility that accepts the same input type as
validate_dir_exists, verifies Path(...).exists() and Path(...).is_file(), and
raises the same error semantics as the directory validator on failure, then
change the call in the class from validate_dir_exists(self.metadata_file_path)
to validate_file_exists(self.metadata_file_path) (or directly use
Path(self.metadata_file_path).is_file() with identical error handling if you
prefer not to add a helper).
- Around line 103-104: The code incorrectly calls validate_dir_exists on
columns_file_path (a file, not a directory); replace that call with the
file-level check (e.g., validate_file_exists(self.columns_file_path)) or
otherwise validate that self.columns_file_path is an existing file (or use
os.path.isdir/os.path.isfile) inside the same class/method where
columns_file_path is set so the validator matches the resource type; update the
reference to validate_dir_exists to the appropriate file validator for
columns_file_path.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@integration-tests/tests/utils/classes.py`:
- Around line 98-99: The code reads JSON metadata with
self.metadata_file_path.read_text() which uses the system default encoding;
change the read to explicitly use UTF-8 (e.g., call read_text(encoding="utf-8")
or open the path with encoding="utf-8") before passing the string to
IntegrationTestDatasetMetadata.model_validate_json so self.metadata is parsed
consistently across platforms; update the read that assigns raw_metadata and
keep the subsequent
IntegrationTestDatasetMetadata.model_validate_json(raw_metadata) call unchanged.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@integration-tests/tests/data/README.md`:
- Line 48: The markdown line uses straight (regular) quotes instead of inline
code formatting for boolean literals; update the text so the boolean values are
wrapped in backticks (e.g., replace "True" and "False" with `True` and `False`)
in the README entry that currently reads '`True` if logs are unstructured, else
`False`.' to ensure proper inline code formatting for the boolean values.

In `@integration-tests/tests/utils/classes.py`:
- Around line 103-114: The metadata validation currently accepts absolute paths
or '..' traversal from metadata (e.g., logs_subdir and file_names) which can
escape the dataset root; update the validation after
validate_dir_exists(self.logs_path) and before calling validate_file_exists to
resolve candidate paths (use Path.resolve or os.path.realpath) for both the
logs_subdir-derived path and each file_path_abs, then assert the resolved path
is contained within the dataset root (self.logs_path.resolve()) — if not, fail
the test (pytest.fail) with a clear message; reference functions/attributes:
validate_dir_exists, validate_file_exists, self.logs_path,
self.metadata.file_names, and any logs_subdir handling so the containment check
covers both directory and file entries.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@integration-tests/tests/data/README.md`:
- Around line 73-83: The fixture example incorrectly passes dataset_name to the
IntegrationTestDataset constructor even though IntegrationTestDataset derives
its name from metadata; update the dataset_name fixture to only construct
IntegrationTestDataset with the path (e.g. return
IntegrationTestDataset(path_to_dataset_root=integration_test_path_config.test_data_path
/ dataset_name)) and keep the fixture name/IntegrationTestPathConfig usage the
same so the runtime won't fail due to an unexpected keyword argument.