fix(realsense): ignore invalid frames for disabled streams

[bowenszhu]

Type / Scope

• Type: Bug
• Scope: RealSense cameras

Summary / Motivation

pyrealsense2 can return a non-None frame object whose boolean value is false when a stream is disabled. SharedRealSenseManager._build_frame_snapshot() previously treated such handles as valid and could call get_data(), raising null pointer passed for argument "frame_ref". This change uses the SDK frame's truthiness so invalid color and depth handles are ignored while valid frames keep the existing decode and copy behavior.

Related issues

• Fixes / Closes: N/A
• Related: N/A

What changed

• Check color and depth frames with truthiness before decoding or reading data.
• Add isolated regression tests for falsey color and depth frame handles.
• Verify valid frame behavior still decodes RGB and copies depth data.
• No breaking changes.

How was this tested (or how to run locally)

• Added tests/cameras/test_realsense_frame_snapshot.py.

• Ran the focused regression tests and dependency check:

  python -m unittest discover -s tests/cameras -p 'test_realsense_frame_snapshot.py' -v
  python -m pip check

  Result: all 3 tests passed, and pip check reported no broken requirements.

• Ran the read-only hardware regression check:

  lerobot-find-cameras realsense \
    --output-dir /tmp/lerobot-camera-regression-check \
    --record-time-s 3

  Detected D435i 261822072940 and D405 260522274196, saved an RGB image from each camera, and observed no frame_ref errors or frame timeouts.

• The existing lerobot Conda environment did not include pytest or Ruff, so those runners could not be executed. No dependencies were installed or upgraded.

Checklist (required before merge)

• Linting/formatting run (pre-commit run -a)
• All tests pass locally (pytest)
• Documentation updated
• CI is green

Reviewer notes

• Please focus on the falsey-but-non-None frame semantics returned by pyrealsense2 for disabled streams.
• The new regression tests use mock frames and do not require a physical RealSense camera.
• Anyone in the community is free to review the PR.