feat(annotate_subtasks): 448x448 frames; select camera0 from standard mapping

[WilliamYue37]

What this does

Two changes to src/opentau/scripts/annotate_subtasks.py (🗃️ Feature):

1. Default frame resolution → 448 × 448 square. Replaces --target-width 640 with --target-size 448. Frames are now scaled so the shorter side equals target_size, then center-cropped to target_size × target_size, matching standard VLM preprocessing. Already-square inputs at the target resolution pass through untouched.

2. camera0 selection from the standard data format mapping. The dataset's camera0 video feature is resolved with this priority:

   1. DatasetConfig.data_features_name_mapping["camera0"] inline on the mixture config (src/opentau/configs/default.py:101),
   2. DATA_FEATURES_NAME_MAPPING[repo_id]["camera0"] from src/opentau/datasets/standard_data_format_mapping.py,
   3. Fallback: first feature in info.json with dtype=='video' (the script's pre-existing behavior).

   Datasets whose camera0 resolves to a feature that is missing from info.json or has dtype != 'video' (e.g. the dummy / vsr / cocoqa rows that map camera0 → "image") are skipped with a warning — they are not LeRobot video datasets.

The module docstring and CLI --help text are updated to match.

How it was tested

• Added TestResizeAndCenterCrop (5 cases: landscape / portrait / already-square / upscale / non-default size) and TestResolveCamera0VideoKey (8 cases: inline-vs-global precedence, fallback paths, non-video feature, missing feature, missing repo_id, no-video-features) in tests/scripts/test_annotate_subtasks.py.
• All 46 tests pass:

  pytest tests/scripts/test_annotate_subtasks.py
  

• Pre-commit (ruff lint+format, pyupgrade, bandit, license header, etc.) all green on both changed files.

How to checkout & try? (for the reviewer)

pytest -sx tests/scripts/test_annotate_subtasks.py

# Dry run on one episode of a dataset whose mixture config has either an
# inline `data_features_name_mapping` or a repo_id present in
# DATA_FEATURES_NAME_MAPPING:
python src/opentau/scripts/annotate_subtasks.py \
    --config-path configs/examples/train_mixture_config.json \
    --max-episodes-per-dataset 1 \
    --no-write-response-column

Checklist

• I have added Google-style docstrings to important functions and ensured function parameters are typed.
• My PR includes policy-related changes.
  • If the above is checked: I have run the GPU pytests (pytest -m "gpu") and regression tests.

🤖 Generated with Claude Code

[claude]

Reviewed — no inline findings. Summary in PR comment.

[claude]

[claude-review] summary for commit 6e0380c

No blocking issues found.