feat: add transform-driven column updates

[FANNG1]

Closes #84.

Summary

• add a transform-driven update_columns API that does not require a prepared DataFrame
• support Lance SQL expressions, BatchUDF, and row-preserving Arrow callables
• support an optional where predicate evaluated before the transform
• schedule work by fragment ID without grouping or shuffling user data
• allow transform workers to request Daft/Ray CPU, GPU, process, and custom Ray resources
• commit rewritten fragments atomically with the existing RewriteColumns transaction path

Correctness

• pin every worker and the final commit to the same Lance snapshot
• validate predicates, target columns, transform schemas, row counts, nullability, and safe casts
• keep no-match updates as no-ops and preserve OCC conflict detection
• request matched physical offsets with fragment.update_columns(..., with_offsets=True) and propagate them through updated_fragment_offsets, including for stable-row-ID datasets
• allow transform-driven updates of a real business column named fragment_id while preserving the grouping-key restriction in update_columns_df
• always begin writes from the current snapshot; the update APIs do not accept version, asof, or default_scan_options

Tests

• uv run pytest tests/io/lance/test_update_columns.py -q (40 passed)
• uv run pytest tests/ -q (492 passed, 10 skipped, 2 xfailed, 2 xpassed)
• Ruff format and lint passed
• focused mypy checks for changed files passed

[FANNG1]

@rchowell @universalmind303 could you take a look when you get a chance? This is the transform-driven half of the column-update work (#84), built on the same RewriteColumns commit path as #58.

[rchowell]

nit: it's a bit odd to have concurrency checks in this method