feat(impact): whole-project value track credited from full fallow runs

[BartWaardenburg]

Summary

• Add a separate whole-project Impact track (project_records) recorded best-effort at the end of a full fallow run, kept apart from the changed-file audit track so the two scopes never share a trend.
• Introduce Scope::ChangedFiles | WholeProject attribution: audit credits changed-file fixes; a full run additionally credits duplication and whole-repo cleanups fixed outside a changed-file audit (scope is the union of frontier keys plus this-run finding and clone paths).
• Gate the project writer airtight in is_whole_project_run: it records only on a true whole-repo run (check + dupes + health, no --changed-since / --workspace / --changed-workspaces, no active diff filter, no production mode). A reshaped clone that is still duplicated under a new fingerprint is not falsely credited resolved.
• ImpactReport gains optional project_surfacing / project_trend (additive; report schema_version stays 1). Human and markdown render an understated whole-project section labelled as whole-repo context, not a to-do, with an "advances only on local full runs, not CI" caveat.
• Regenerate the MCP impact tool description, JSON schema, and TS contracts for the new fields.

Test plan

• cargo test --workspace --all-targets: zero failures (pre-push gate green).
• New render-state unit tests cover project-only and both-tracks rendering in human + markdown.
• Attribution safety tests (resolved / suppressed / reshaped-clone) use real temp files so frontier pruning is exercised.
• cargo clippy --workspace --all-targets -- -D warnings clean; cargo fmt --all -- --check clean.
• OLD vs NEW smoke: a full fallow run accrues the project track; --only dupes, --changed-since, --production, and --diff-stdin all keep project_records empty; the actionable-vs-context labels render correctly in the release binary.