Split finalization write and cleanup profile diagnostics

[jununfly]

This was generated by AI during triage.

Parent

Plan: docs/plans/2026-06-20-rust-hybrid-finalization-cleanup-diagnostics-and-batching.md

What to build

Add more specific profile diagnostics for the TypeScript-owned finalization write/cleanup tail while preserving the existing high-level profile fields. The goal is to make edge materialization, endpoint validation, edge insert, resolved-reference cleanup, intentionally-unresolved cleanup, and related row/edge counts visible enough for later optimization decisions.

Do not change graph semantics, reference disambiguation, SQLite schema, CLI/SDK/MCP behavior, or README metrics. This slice is diagnostics only.

Acceptance criteria

• Existing high-level fields such as edgeMaterializationMs, edgeWriteMs, unresolvedCleanupMs, and databaseAccessMs remain present.
• Profile diagnostics expose sub-buckets or equivalent fields that answer: endpoint validation time, edge insert time, resolved cleanup time, intentionally unresolved cleanup time, inserted edge count, resolved cleanup row count, and intentionally unresolved cleanup row count.
• A deterministic profile-shape test covers the new fields.
• Existing profile/status/doctor consumers continue to pass.
• No schema, disambiguation, import/framework, dynamic-dispatch, or user-facing behavior changes.

Blocked by

None - can start immediately

[jununfly]

This was generated by AI during triage.

Implemented. Finalization profile diagnostics now preserve the existing high-level buckets and add sub-buckets/counts for edge endpoint validation, edge insert count, resolved cleanup, and intentionally unresolved cleanup. Covered by the rust-hybrid profile-shape test.

[jununfly]

This was generated by AI during triage.

Closing as implemented: profile diagnostics now expose finalization write/cleanup sub-buckets while preserving existing high-level fields.