Broad Read-Write test

Want a broad test that could catch issues like: https://github.com/lancedb/lance/pull/4396

Can do similar plan to https://github.com/lancedb/lance/issues/4416

• Have one large master schema with all primitive types plus common nested ones
• Have all index types on same dataset, including common variations
  • all distance types for IVF PQ
  • with_positions for FTS

Then vary the state of that dataset:

• Fragmentation: one fragment, two rows per fragment
• Deletions: No deletions, alternating deletions
• Index coverage: full coverage, partial coverage

That gives 6 datasets to test.

Then run:

• Full scan
• Query that hits each index (with varying queries)
• Append
• Optimize each index