Improve language server responsiveness

[renkun-ken]

Summary

• Precompute semantic tokens and symbol/call indexes in background parse workers.
• Add semantic-token delta responses, request cancellation/coalescing, and correct range filtering.
• Use indexed lookups for references, rename, highlights, code lenses, and call hierarchy.
• Prioritize parse work, supersede stale tasks, keep warm asynchronous workers, and defer diagnostics.
• Bound parse and diagnostics caches by memory and constrain range providers to requested lines.

Performance

• 3,000-line semantic-token full response: ~813 ms to ~1 ms.
• 100-line inlay-hint request: ~3.3 s to ~130 ms.
• References with 300 uses: ~2.4 s to ~10-40 ms.
• 3,000-line parse: ~950 ms to ~725 ms despite the additional indexes.

Validation

• Full test suite passed.
• R CMD check --no-manual passed; the only warning is a local Apple compiler flag warning.
• git diff --check passed.