fix: #430 HNSW insert beam + distance-based pruning + storage rebuild + CI guards

[ruvnet]

Summary

Closes the three remaining root causes from #430 plus the storage-rebuild gap that PR #460 separately identified. All previously-undocumented HNSW correctness bugs at scale.

• ruvector-router-core HNSW: search returns 0 results from inserted vectors at scale (recall@1 << 1.0) #430 Bug B — insert beam clamp. ef_construction.min(m * 2) silently became m * 2 (32 with defaults) instead of 200. Late-inserted clusters got wired through whatever sat near the entry point. Removed the clamp.
• ruvector-router-core HNSW: search returns 0 results from inserted vectors at scale (recall@1 << 1.0) #430 Bug C — FIFO adjacency pruning. Overflow trim used drain(0..drain_count), dropping the OLDEST edges regardless of distance. Proper HNSW keeps the m closest. Now scores via calculate_distance and truncates to m, with a guarded fallback if the anchor vector lookup fails.
• ruvector-router-core HNSW: search returns 0 results from inserted vectors at scale (recall@1 << 1.0) #430 Storage rebuild. VectorDB::new() always created a fresh empty HnswIndex, so persisted vectors were invisible to search after reopen. Rebuild via storage.get_all_ids() + index.insert_batch() on open and seed total_vectors from the recovered count.

Tests

• test_pruning_keeps_closest_not_newest — builds a hub with 20 close neighbours then 6 far neighbours, asserts no far_* id appears in top-10 around the hub. Fails on FIFO pruning.
• test_index_rebuilt_from_storage_on_open — writes 5 vectors with one VectorDB, reopens against the same path, asserts search returns the persisted match. Fails on the empty-index regression.

All 21 router-core unit tests pass locally (release):
```
test result: ok. 21 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
```

Regression-guard CI

Three new structural guards complement the behavioural hnsw-recall-at-1 job:

Guard	What it forbids
`hnsw-insert-beam-no-m2-clamp`	textual `ef_construction.min(m * 2)` in `index.rs`
`hnsw-distance-based-neighbor-pruning`	requires `calculate_distance` + the `> m * 2` overflow gate to coexist
`vector-db-rebuilds-index-on-open`	requires `storage.get_all_ids()` in `vector_db.rs`

The existing `hnsw-recall-at-1` job now also runs `test_pruning_keeps_closest_not_newest` and `test_index_rebuilt_from_storage_on_open`.

Release plan

This PR does NOT bump `@ruvector/router` — the original 0.1.31 bump was reverted because the `optional-deps-resolvable-on-npm` regression guard fails when platform-specific subpackages don't yet exist on npm (CI publishes them only after a tag is cut).

After this PR merges, a separate release PR will:

1. Re-bump @ruvector/router meta + 5 platform subpackages 0.1.30 → 0.1.31.
2. Tag v0.1.31 (or workflow_dispatch on `publish-all.yml`) so CI builds + publishes all platform binaries together.

Test plan

• CI green on `regression-guard.yml`
• CI green on `ci.yml` (build + test)
• After merge: release PR bumps router 0.1.30 → 0.1.31 → tag → CI publishes
• Post-publish: `npx @ruvector/router@0.1.31` smoke test on the recall-at-1 repro

Supersedes

PR #460 (CoolDude1969) — overlapping work; this PR keeps the storage-rebuild idea and adds bugs B + C + CI guards. Credited in the closing comment.

🤖 Generated with claude-flow