Apply linting to entire codebase & add CI workflow to check linting - #3216

Introduce tools/ruff_report.py: a CLI script that runs `ruff` on given paths (using `--output-format=json --exit-zero`), parses the JSON results, and generates a readable Markdown report. The report groups issues by rule and file, includes a summary table, suggested triage order, table of contents, per-rule sections with detailed issue tables, and VS Code "Quick open" commands. The script also embeds short RULE_NOTES for common codes, accepts paths (default `.`) and `--output` (default `ruff-report.md`), and writes the generated Markdown to disk.

Add tools/trim_lines.py: a command-line helper that finds files with Ruff E501 violations and attempts to reduce them using autopep8, then normalizes formatting with Ruff. It scans paths via `ruff check --output-format=json`, supports batching (50 files), a configurable `--line-length` (default 88), and a `--check` dry-run mode. The script prints invoked commands and outputs, runs autopep8 with selected fixes, then runs `ruff --fix --unsafe-fixes` and `ruff format`. Requires ruff and autopep8.

Standardize docstring punctuation and formatting across modules, add stacklevel=2 to warnings.warn calls for clearer warning locations, and make zip/dict(zip()) usages more robust by passing strict=False where appropriate. Also include minor loop variable/name cleanups and CLI help text formatting fixes. These changes improve error tolerance when iterables differ in length, produce more helpful warning tracebacks, and tidy user-facing documentation strings.

Delete the generated ruff-report.md (large auto-generated lint report) and add tmp/* to .gitignore to avoid committing tool output. Also include an update to tools/ruff_report.py (script changes) alongside the cleanup.

Rework deeplabcut package initialization: use relative imports, add a logger and robust DEBUG env parsing, and replace eager imports of optional GUI/PyTorch features with a lazy-loading mechanism. Optional public names (GUI and transformer_reID) are resolved on first access via __getattr__ (with clear error messages when dependencies are missing) and cached for subsequent use; __dir__ and a consolidated __all__ improve discoverability. Removes previous noisy prints and light-mode fallbacks in favor of structured logging and explicit public API grouping.

Change get_fmpose3d_inference_api to use None as the default for config_kwargs and initialize it to an empty dict inside the function. This avoids the shared mutable default argument ({}), preventing accidental state leakage across calls. Updated in deeplabcut/pose_estimation_pytorch/modelzoo/fmpose_3d/fmpose3d.py.

Small refactor in fmpose_3d inference: bind outer variables (predictions_2d, all_poses_3d) as default args for the nested _process_batch to ensure correct capture when used as a callback, and tidy up several long-wrapped statements for readability (create_empty_df call, dest_folder handling, logger.info, and the poses_3d_serialisable list comprehension). No functional changes intended.

Clean up pyproject.toml by removing the temporary [tool.yapf] linting configuration and its related comment. This deletes the previous based_on_style and indent_width settings no longer needed after linting changes.