feat: report index coverage, segment count and size in SHOW INDEXES

[ivscheianu]

Part of #789.

What

SHOW INDEXES reports indexed and unindexed fragment and row counts, but leaves the operator to compute coverage by hand and says nothing about how the index is laid out or what it costs. This adds three columns:

Column	Type	Description
indexed_percent	double	Share of rows covered, truncated to two decimals so it never overstates. Null for an empty table, where coverage is undefined rather than complete.
num_segments	long	Physical segments backing the logical index.
size_bytes	long	Total across the segments' files. Null if any segment predates index file size tracking, since a partial sum would understate rather than admit it is unknown.

All three come from metadata the command already fetches, so there is no extra driver cost. Truncating rather than rounding matters: a table one row short of full coverage reads as 99.99, not 100.

Change

indexed_percent is derived from the row counts already read. size_bytes needs the whole segment list rather than its first element, so the grouping keeps it instead of calling .head immediately; output is still one row per logical index. num_segments reads the statistics blob, falling back to num_indices for older cores that report only that key.

The docs note two things an operator needs and would otherwise guess wrong:

• Partial coverage is not merely slower. A partially covered zonemap index prunes the fragments it does not cover, so a predicate on the indexed column can return fewer rows than the table holds, while COUNT(*) over the table stays correct. It comes from the connector's own driver-side pruning, which builds its surviving fragment set purely from the fragments the committed zones name, with no gate on coverage. Reproducible with CREATE INDEX plus an append, so it predates this change, but this column is what makes it visible. btree, bitmap and bloomfilter return complete results while partially covered.

  Edit: fixed by fix(read): keep fragments the zonemap does not cover #781.

• A vector index is listed here too, but Spark SQL can neither create nor refresh one; it is maintained through the Lance SDK.

[lance-gatekeeper]

✅ Gate recommendation: approve.

The latest cleanup is behavior-preserving: the inferred boxed values and imported test symbols compile under both Scala 2.12 and 2.13. The implementation still derives coverage from existing metadata, preserves one row per logical index, and reports NULL rather than understating unknown size; #781 remains merged, so the documentation correction is also current.