zipper,node: reduce wrapper churn and grow key scratch geometrically

[snissn]

Summary

• switch zipper loadNode to node.NewNodeView/value wrappers to avoid per-load *Node allocations
• thread value-node wrappers through recursive/coalesce paths
• grow node.ensureKeyScratch geometrically to avoid realloc churn on small key-length increases
• add node unit coverage for scratch growth reuse behavior

Validation

• go test ./TreeDB/node ./TreeDB/zipper
• go test ./TreeDB/caching ./TreeDB/internal/memtable ./TreeDB/zipper ./TreeDB/node
• unified bench: /home/mikers/tmp/perf-pr3only-1773439188 (base: /home/mikers/tmp/perf-base820-1773438485)
  • seq write: 4,160,181 vs 4,133,530
  • random write: 5,076,820 vs 5,105,302
  • batch_write_steady alloc_space writeRecursive: 47.69MB vs 48.69MB

[Copilot]

Pull request overview

This PR reduces allocations in hot read/merge paths by switching zipper node loading to use node.Node value views (via NewNodeView) and by improving Node’s internal key-scratch growth strategy to avoid repeated small allocations during leaf key reconstruction.

Changes:

• Update Zipper.loadNode to return node.Node values and use node.NewNodeView, adjusting downstream call sites to pass *node.Node where required.
• Change (*node.Node).ensureKeyScratch to grow geometrically (with overflow guard) instead of reallocating to the exact requested size.
• Add a focused unit test validating key-scratch geometric reuse behavior.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated no comments.

File	Description
TreeDB/zipper/zipper.go	Switch node loading from *node.Node to allocation-free node.Node views and update call sites accordingly.
TreeDB/node/leaf.go	Make ensureKeyScratch grow geometrically (min 64 bytes) to reduce repeated allocations.
TreeDB/node/key_scratch_test.go	Add test ensuring small size increases reuse capacity and larger increases trigger growth.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

This PR reduces heap allocations and improves scratch-buffer reuse in TreeDB’s page/node handling, primarily around zipper node loading and leaf key decoding.

Changes:

• Update Zipper.loadNode to return node.Node by value and use node.NewNodeView (no heap allocation) for loaded pages.
• Adjust zipper merge/coalesce call sites to pass *node.Node where required after the value-return change.
• Improve node.Node.ensureKeyScratch to grow geometrically and add a unit test validating reuse behavior.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated no comments.

File	Description
TreeDB/zipper/zipper.go	Switch node loading to allocation-free Node values and update merge/coalesce call sites accordingly.
TreeDB/node/leaf.go	Change key scratch allocation strategy to geometric growth to avoid repeated small allocations.
TreeDB/node/key_scratch_test.go	Add coverage to ensure key scratch capacity is reused on small growth and increases when necessary.
TreeDB/caching/db.go	Formatting/alignment-only change within arena sizing constants.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.