[codex] Extract Rerun reader

[Guilherme Penedo (guipenedo)]

This PR extracts the Rerun reader into its own branch and keeps the reader-only surface separate from the writer, cleanup, and benchmark work.

Included here:

• read_rerun pipeline entrypoints and package exports
• the shared _rerun_io.py helper used by the reader
• reader-only docs and optional-dependency updates
• the reader test suite
• the minimal core pipeline fixes needed for reader projections and metadata-only batch handling

Excluded from this branch:

• write_rerun
• Rerun writer sink code
• cleanup/reducer changes
• benchmark harnesses and performance-only tuning

Validation:

• uv run ruff check ... on the reader branch files
• uv run ty check ... on the reader branch files
• uv run pytest tests/readers/test_rerun_reader.py

[gemini-code-assist]

Code Review

This pull request introduces a new Rerun reader (read_rerun) to parse .rrd files into columnar recording rows or robotics episode rows, along with corresponding documentation, tests, and dependency updates. Feedback on the changes highlights two critical issues: first, a deserialization bug in LocalRrd.__setstate__ where restoring local paths for remote sources can cause FileNotFoundError in distributed environments; second, a potential crash in _iter_encoded_images when handling empty or null image slices. Both issues include actionable code suggestions to resolve them.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

In __setstate__, restoring self.path from the serialized state when the source is a remote file (i.e., self.source.is_local is False) will cause subsequent calls to open() on the deserialized machine to return a non-existent local temporary path from the serialization machine. This will lead to a FileNotFoundError during distributed or cloud execution. Only restore self.path if the source is local.

Suggested change

	def __setstate__(self, state: dict[str, object]) -> None:
	self.source = cast(DataFile, state["source"])
	self.tmpdir = None
	path = state.get("path")
	self.path = Path(path) if isinstance(path, str) else None
	def __setstate__(self, state: dict[str, object]) -> None:
	self.source = cast(DataFile, state["source"])
	self.tmpdir = None
	path = state.get("path")
	self.path = Path(path) if isinstance(path, str) and self.source.is_local else None

[gemini-code-assist]

If the inner image list is empty or null (e.g., a missing frame on the timeline), inner_offsets[outer_start + 1] will be equal to or less than inner_offsets[outer_start]. This results in byte_end <= byte_start, which causes inner_values.slice to return an empty array, leading to a crash in PIL.Image.open with UnidentifiedImageError. Add a guard to skip empty/null image slices.

Suggested change

	byte_start = int(inner_offsets[outer_start])
	byte_end = int(inner_offsets[outer_start + 1])
	data = np.asarray(
	inner_values.slice(byte_start, byte_end - byte_start)
	).tobytes()
	byte_start = int(inner_offsets[outer_start])
	byte_end = int(inner_offsets[outer_start + 1])
	if byte_end <= byte_start:
	continue
	data = np.asarray(
	inner_values.slice(byte_start, byte_end - byte_start)
	).tobytes()