Bound session parsing concurrency and report worker panics per file

[coderabbitai]

Summary

Bound session parsing concurrency in load_session_files and preserve per-file reporting when a worker panics.

Rationale

The current implementation in crates/sivtr-core/src/query/mod.rs creates one scoped thread per session file. Workspaces with many session files can oversubscribe CPU resources, increase peak memory use, or reach host thread limits. A panic during JoinHandle::join() can also stop the full load instead of reporting the affected session as skipped.

Required changes

• Limit parsing workers to std::thread::available_parallelism(), with a safe fallback of one worker.
• Distribute session-file tasks across the bounded worker set.
• Continue processing other session files if one worker panics.
• Mark every task from a panicking worker as a SkippedSession with the error text parse worker panicked.
• Do not use unwrap() in production code.

Affected area

• crates/sivtr-core/src/query/mod.rs
• load_session_files

Acceptance criteria

• The number of parsing threads does not exceed the selected worker limit.
• Normal session-file outcomes retain their existing ordering.
• A worker panic does not abort the complete workspace load.
• Each session file not completed because of a worker panic appears in QueryResult::skipped.
• Relevant tests and strict Clippy checks pass.

Backlinks

• Deferred from PR feat(search): incremental listing cache and light/full session views #179: feat(search): incremental listing cache and light/full session views #179
• Review discussion: feat(search): incremental listing cache and light/full session views #179 (comment)
• Requested by @Ariestar.