Refactoring the Demuxers for chunked Tensor data

[rwkeane]

This commit refactors TensorDemuxer and SmoothedTensorDemuxer to process SerializableTensorChunk objects, aligning with a chunk-based data model. Unit tests have been carefully updated to ensure minimal diffs for existing test cases while adding new tests for chunk-specific scenarios.

Core Refactoring:

1. TensorDemuxer (tensor_demuxer.py):

   • I renamed on_update_received to on_chunk_received, accepting
     SerializableTensorChunk (using chunk.tensor for data and
     chunk.timestamp.as_datetime() for timestamp conversion).
   • I preserved internal keyframe storage and cascading update logic.
   • I ensured _on_keyframe_updated is called at most once per affected
     timestamp after a chunk is processed and causes a state change.

2. SmoothedTensorDemuxer (smoothed_tensor_demuxer.py):

   • I updated on_chunk_received to delegate to the parent class's
     method for chunk processing.

Unit Test Updates (Iterative Approach):

• Initial Broad Refactor & Subsequent Refinement: I iteratively refined test updates based on feedback to ensure minimal changes to
  existing test cases.
• tensor_demuxer_unittest.py:
  • Existing tests: Calls to on_update_received(idx, val, ts) were
    transformed to on_chunk_received(Chunk(ts, idx, [val])).
  • Assertions for mock client notifications (call_count) were
    carefully adjusted to reflect the rule that notifications occur
    only if a keyframe's final calculated value changes, for both
    direct updates and cascades.
  • I added new tests to specifically cover scenarios with
    multi-update chunks.
• smoothed_tensor_demuxer_unittest.py:
  • Minimal changes were required as these tests primarily verify
    smoothing logic by directly manipulating parent keyframe history,
    not by calling the data input method.
• Test Helpers: I added MockSynchronizedTimestamp and an inlined
  SerializableTensorChunk (with tensor field) to test
  files to ensure compatibility with TensorDemuxer's expectation of
  timestamp.as_datetime() and field naming.
• E2E Test Adjustments: I applied necessary fixes to
  tensor_e2etest.py for compatibility with SerializableTensorChunk
  (field name and timestamp object).

Code Quality and Verification:

• Comments: I reviewed and refined application code comments to
  explain "why" not just "what," and removed placeholder/meta-comments.
• Placeholder File: I deleted the temporary
  serializable_tensor_chunk_placeholder.py.
• Static Analysis: All changes passed Black, Ruff, MyPy, and Pylint.
• Testing: The full pytest suite (895 tests) passed consistently
  across two runs, confirming the correctness and integration of changes.

This work completes the transition of the demuxer components to the new chunk-based data transport model while maintaining test coverage and adhering to project standards.