Compaction: eliminating disk scan and updating PQ retrain

[dian-lun-lin]

computeLayerInfoFromSources called getNodes(0) on each source graph to count live nodes at level 0. getNodes(0) sequentially seeks through every node record on disk to filter out deleted entries. On a cold page cache this touches large amounts of source data before compaction even begins, significantly delaying the start of actual graph merging.

Since every live node is present at level 0 by the HNSW invariant, the count is simply liveNodes.get(s).cardinality() — an in-memory popcount requiring no I/O.

Also switch PQ retraining from ProductQuantization.compute() (full k-means++ init) to basePQ.refine() (Lloyd's iterations only, warm-started from the existing codebook). The source codebooks are already trained on the same distribution, so warm-starting converges in far fewer passes with no recall loss.

Dataset	Workload	main cold	fix cold	Δ cold	main recall	fix recall
cohere-1M	2-TIERED	2m56s	1m07s	−62%	0.6099	0.6091
cohere-1M	2-UNIFORM	2m58s	1m05s	−63%	0.6199	0.6225
cohere-1M	4-UNIFORM	3m06s	1m10s	−63%	0.6247	0.6257
cohere-10M	2-TIERED	31m12s	13m37s	−56%	0.5647	0.5681
cohere-10M	2-UNIFORM	27m23s	6m31s	−76%	0.5665	0.5699
cohere-10M	4-UNIFORM	32m40s	7m14s	−78%	0.5698	0.5714
cap-1M	2-TIERED	3m15s	1m20s	−59%	0.6672	0.6645
cap-1M	2-UNIFORM	3m15s	1m17s	−60%	0.6683	0.6701
cap-1M	4-UNIFORM	3m50s	1m32s	−60%	0.6739	0.6715
cap-6M	2-TIERED	17m52s	4m28s	−75%	0.6303	0.6307
cap-6M	2-UNIFORM	17m39s	4m16s	−76%	0.6375	0.6346
cap-6M	4-UNIFORM	19m02s	5m44s	−70%	0.6405	0.6407
dpr-gemma-1m	2-TIERED	2m49s	56s	−67%	0.7794	0.7810
dpr-gemma-1m	2-UNIFORM	2m46s	49s	−71%	0.7770	0.7777
dpr-gemma-1m	4-UNIFORM	2m55s	1m06s	−63%	0.7858	0.7858
dpr-gemma-10m	2-TIERED	25m55s	4m03s	−84%	0.6932	0.6928
dpr-gemma-10m	2-UNIFORM	26m01s	4m11s	−84%	0.6893	0.6947
dpr-gemma-10m	4-UNIFORM	27m52s	5m45s	−79%	0.7048	0.7061

[github-actions]

Before you submit for review:

• Does your PR follow guidelines from CONTRIBUTIONS.md?
• Did you summarize what this PR does clearly and concisely?
• Did you include performance data for changes which may be performance impacting?
• Did you include useful docs for any user-facing changes or features?
• Did you include useful javadocs for developer oriented changes, explaining new concepts or key changes?
• Did you rebase your branch onto the latest main for regression testing and PR submission?
• Did you trigger regression testing via Run Bench Main and review results?
• Did you adhere to the code formatting guidelines (TBD)
• Did you group your changes for easy review, providing meaningful descriptions for each commit?
• Did you ensure that all files contain the correct copyright header?
• Did you add documentation for this feature to the release notes directory?

If you did not complete any of these, then please explain below.