Reuse valid search indexes without delaying notebook loading

[emsi]

Summary

• reuse a published search-index generation only when source identity, fingerprint, schema, projection version, and projection profile all match
• move source loading and serialized index mutation onto independent workers with source-scoped generations and cancellation
• report reused, rebuilt, failed, and pending index work without exposing notebook content
• preserve notebook presentation order independently from restored-source load priority

Motivation

Startup previously deleted and rebuilt every successfully loaded notebook's search documents, even when the source and indexed projection were unchanged. Indexing also ran on the source-loading worker, so rebuilding one notebook delayed loading the next.

The optimized startup initially prioritized the last active notebook correctly but accidentally allowed load completion order to become navigation order. This PR keeps the restored source first in the work queue while inserting completed notebooks into their persisted presentation positions.

Validation

• cargo test --workspace
• cargo clippy --workspace --all-targets -- -D warnings
• scripts/validate-docs.sh
• isolated native GUI smoke test
• GitHub Actions release workflow, including native tests, Flatpak build and bundled-extractor smoke test, and AppImage build and smoke test: https://github.com/emsi/OneNoteViewer/actions/runs/31964874546
• regression tests verify unchanged index reuse, transactional replacement, stale-work cancellation, non-blocking loading, persisted source order, and GTK positional insertion

Closes #54
Related to #37
Builds on #38