fix: insert/insert_batch now write to KV store; fix BM25 fallback to …

…use proper IDF formula

Three bug fixes in namespace.py:

1. insert() now dual-writes to HNSW (in-memory) AND KV store
   - Before: insert() only populated _vector_index and _metadata_store
   - After:  also persists JSON doc to KV using same schema as insert_multi()
   - Impact: docs inserted via insert() were invisible to keyword_search()
             and the BM25 leg of hybrid_search()

2. insert_batch() now dual-writes to HNSW AND KV store
   - Same root cause as insert() — batch inserts were KV-invisible
   - All docs in a batch are written in a single atomic transaction

3. Python BM25 fallback replaced TF-count with proper BM25 formula
   - Before: score += count(term in doc)  — no IDF, no length normalisation
   - After:  BM25 with k1=1.2, b=0.75 (Robertson-Sparck Jones IDF)
   - Matches the native Rust bm25.rs implementation (bm25.rs: k1=1.2, b=0.75)
   - Fixes ranking quality when FFI keyword search returns None

Bumps version to 0.5.5.

bump version to 0.5.4

Added missing features.

bump version to 0.5.2

Merge pull request #15 from sochdb/release/0.5.0

Performance tweaks.

Merge pull request #15 from sochdb/release/0.5.0

Performance tweaks.

docs: move Quick Start into Table of Contents structure

- Quick Start is now section #1 after Table of Contents
- Architecture moved before ToC
- Better document structure: Install → Architecture → ToC → Content
- Consistent with other SDKs

Merge pull request #14 from sochdb/release/0.4.8

Release 0.4.8 - Concurrent embedded mode

FIxed lib auto detect.

Bump version to 0.4.6

- Auto-detection improvements for library paths
- Support for Homebrew, MacPorts, user installations
- Better error messages with installation instructions