perf(index): apply the warm-load cache by move — cut the startup RSS peak

[Hessesian]

Summary

F0 of the memory plan: the warm-load apply path held the deserialized WorkspaceIndexResult AND a second copy of every FileData simultaneously (file_contributions cloned into the Indexer's Arcs while result.files still owned the originals), and jemalloc retained the freed transients — users saw ~2× the necessary RSS after startup.

• apply_workspace_result now consumes result by value; file_contributions_owned moves each FileData into its Arc (Arc::new(result.data), no clone).
• The Arc<WorkspaceIndexResult> wrapper is gone — it existed only so send_progress_end could read counts after the spawn; the three counts are captured before the move. Review verified the sequentiality (apply is awaited before progress-end; no other holder), and that the old code never used the Arc for retry either.
• A second candidate (cache_entry_to_file_result copy) was implemented, measured at 0 MB peak effect (both builders free-as-they-go), and reverted — deferred to pair with the interning work.

Measured (the probe in the base PR is the meter)

RSS apply peak: 487.1 → 408.1 MB (−79 MB); unaccounted slack 207.9 → 129.1 MB — the delta matches the eliminated single FileData duplication; the remaining slack is the genuinely-derived FileContributions maps, which the planned file-id interning targets next.

1440 tests green, clippy -D warnings clean. Task-scoped review: Approved (no Critical/Important).

Stacked on the probe PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EJsyQ1UgF8uJYR4HgpxEBB

[Copilot]

Pull request overview

This PR reduces warm-load startup memory by eliminating a second in-memory copy of each FileData during the apply phase. It does this by making apply_workspace_result consume WorkspaceIndexResult and moving each FileData directly into the Indexer’s Arc<FileData>, while still emitting the same progress summary by capturing counts before the move.

Changes:

• Change Indexer::apply_workspace_result to take WorkspaceIndexResult by value and drain result.files, moving FileData into Arc<FileData> via a new file_contributions_owned.
• Remove the Arc<WorkspaceIndexResult> wrapper in the warm-apply path and capture progress summary counts up-front instead.
• Update tests/docs in apply_tests and memory_probe_tests to reflect the new ownership/move behavior and expected apply-phase RSS profile.

Reviewed changes

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

File	Description
src/indexer/scan.rs	Captures progress counts before spawning apply, and passes WorkspaceIndexResult by value into apply_workspace_result to enable move-based apply.
src/indexer/memory_probe_tests.rs	Updates the probe narrative and call site to reflect apply_workspace_result consuming the result (no post-apply drop(result)).
src/indexer/apply.rs	Introduces file_contributions_owned and changes apply_workspace_result to consume and drain result.files, moving FileData into Arc without deep clones.
src/indexer/apply_tests.rs	Updates tests to pass WorkspaceIndexResult by value to the new apply_workspace_result signature.