feat(datasets): override + require non-empty robot_type/control_mode

[shuheng-liu]

What this does

Refs #204.

Adds two related knobs for robot_type and control_mode metadata:

• Dataset-level overrides — DatasetConfig.robot_type and
  DatasetConfig.control_mode (both str | None, default None = no
  override). When set, they are written through to dataset.meta.info after
  the dataset is constructed in make_dataset. An explicit "" is treated as
  "clear the loaded value" (distinct from None).
• Mixture-level non-empty requirements — DatasetMixtureConfig.require_non_empty_robot_type
  and require_non_empty_control_mode (both bool, default False).
  Defaults preserve today's behavior: empty values are tolerated. When opted
  into, make_dataset_mixture raises with a per-dataset breakdown of which
  field is missing.

When a control_mode override is provided, the "missing control_mode" warning
in LeRobotDataset.__init__ is pre-suppressed for the affected repo_id —
the user already knew it was missing and is providing a value on purpose.

How it was tested

• Added TestRobotTypeControlModeOverridesAndRequirements in
  tests/datasets/test_dataset_mixture.py (15 tests):
  • DatasetConfig defaults / accepts overrides / accepts explicit "".
  • DatasetMixtureConfig require flags default False.
  • _apply_metadata_overrides: writes through, no-op on None, "" clears,
    skips dataset.control_mode attr update for VQA datasets that don't have
    one.
  • _validate_metadata_requirements: no-op when flags off; passes when
    populated; raises for empty / None / missing values; reports label +
    repo_id; aggregates multiple failures.
• pre-commit run --all-files clean.
• pytest -m "not gpu and not slow" tests/datasets/test_dataset_mixture.py tests/datasets/test_datasets.py tests/datasets/test_optional_keys.py all green.

How to checkout & try? (for the reviewer)

pytest -sx tests/datasets/test_dataset_mixture.py::TestRobotTypeControlModeOverridesAndRequirements

Example config snippet that opts in to the strict checks:

{
  "dataset_mixture": {
    "datasets": [
      {"repo_id": "lerobot/droid_100", "robot_type": "panda", "control_mode": "joint"}
    ],
    "require_non_empty_robot_type": true,
    "require_non_empty_control_mode": true
  }
}

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.

Note: Before submitting this PR, please read the contributor guideline.

---

Generated by Claude Code

[claude]

Inline notes — see summary comment for rollup.

[claude]

[claude-review] summary for commit 27d500a

• suggestion — src/opentau/datasets/factory.py:300 — silent length-mismatch early return bypasses the require-non-empty checks the caller opted into; branch is dead in practice — drop it or convert to assert.
• nit — src/opentau/datasets/factory.py:206 — pre-suppression must run before the LeRobotDataset(...) constructor on the next lines; add a one-line comment about the ordering invariant.
• nit — src/opentau/datasets/factory.py:81 — cross-module import of the private _CONTROL_MODE_WARNED set; consider exposing a public helper in lerobot_dataset.py instead.

[shuheng-liu]

@claude fix per suggestion and nits if they make sense.