[hello-peter-tang]

Summary

Adds daft_lance.read_lance_changes, a lazy Daft scan source over PyLance's
version-range delta readers. Closes #64.

• Modes: "inserted", "updated", and "deleted" (deleted row ids), each
  producing a DataFrame with the upstream result schema: dataset columns plus
  _rowid / _row_created_at_version / _row_last_updated_at_version for
  inserted and updated rows, _rowid only for deleted rows.
• Both endpoints are opened and validated on the driver, and the end snapshot
  is pinned in the worker open kwargs, so a commit after planning cannot change
  either endpoint or the result.
• The dataset and Arrow RecordBatchReader are constructed and consumed on
  scan workers, yielding one batch at a time rather than collecting on the
  driver.
• Explicit validation: begin_version >= 1, begin_version <= end_version,
  existing endpoints (the begin manifest is opened too), and stable row ids at
  both endpoints unless the range is empty.
• Deleted results are rows live at the begin version and absent at the end
  version, so insert-then-delete and compaction-only rewrites are not reported
  as deletions.
• Namespace addressing (table_id + namespace_impl + namespace_properties)
  reuses the existing worker open-context reconstruction.

Tests

tests/io/lance/test_lance_delta.py (50 tests) covers:

• Parity with PyLance across append, repeated update, row and whole-fragment
  deletion, and empty results, including explicit schema checks.
• Deleted ids against begin-live − end-live, insert-then-delete,
  delete-then-reinsert, and compaction cases.
• Equal, reversed, missing, and cleaned-up endpoints; non-integer inputs;
  stable-row-id prerequisites; unknown modes.
• Worker-side execution and incremental batch consumption (unit test on the
  factory generator), plus a commit after planning leaving results unchanged.
• Dir-namespace context reconstruction.

DAFT_ANALYTICS_ENABLED=0 uv run pytest tests/ -q → 615 passed, 10 skipped,
2 xfailed, 2 xpassed. ruff format --check, ruff check, and mypy daft_lance/ are clean.

[hello-peter-tang]

Heads-up: the PR Test Suite / PR Checks run for this branch is sitting in action_required, so the PR shows no CI results at all. GitHub is holding fork runs until a maintainer approves them ("Approve and run"). @maintainers, could someone approve the pending run so this gets tested?

[hello-peter-tang]

Same as my note on #88: CI is still parked in action_required, so I ran the checks locally on the PR head 6c778d9:

ruff check daft_lance tests            -> All checks passed!
ruff format --check daft_lance tests   -> 51 files already formatted
mypy --strict --ignore-missing-imports daft_lance
                                       -> Success: no issues found in 16 source files
pytest tests/ -q                       -> 602 passed, 10 skipped, 2 xfailed, 2 xpassed in 30.92s

Versus main on the same setup (552 passed, 10 skipped, 2 xfailed, 2 xpassed) this branch adds 50 passing tests, matching the new test_lance_delta.py cases. Environment: Linux x86_64, Python 3.11.6, deps from the committed uv.lock, DAFT_ANALYTICS_ENABLED=0. The 3.10/3.12/3.13 matrix and the Ray job still need a real CI run.