Add block-based storage for multivectors

[ArnavAgrawal03]

Add configuration toggles and defaults for multivector block storage. Introduce MultiVectorBlock serialization to stage multivectors in block files and update FastMultiVectorStore to write blocks with cache-aware loading and legacy fallbacks. Improve multivector retrieval by batching block loads and normalizing cache keys; tests not run (not requested).

[jazzberry-ai]

Bug Report

Name	Severity	Example test case	Description
Incorrect Chunk Assignment After Block Upload Failure	Medium	1. Configure block storage with block_size=2. 2. Create 3 DocumentChunks. 3. Mock _upload_block to fail for the first block. 4. Call save_multivectors_to_blocks. 5. Verify bucket/key pairs are correctly assigned.	In _flush_block_entries, legacy fallback can cause incorrect bucket/key assignment if block upload fails.

_{Comments? Email us.}