perf: reduce search latency with synchronous UI-thread search and lock-free tree rebuild

[bosiakov]

• Run search + tree rebuild synchronously on the UI thread (~1-3ms) so results appear on the same frame the user types - zero latency
• Refactor tree rebuild into a pure computeTreeRebuild() function using snapshot-compute-swap to minimize lock hold time
• Add Parent pointers to prefix tree nodes and TreeNode back-pointers on commands, enabling O(matches × depth) ancestor marking instead of O(total_nodes) full tree walks
• Optimize hot path: stack-allocate Levenshtein buffers for short strings, add fast determineMatchTypeShort path for 1-2 char queries, cache TreeNodesCount at build time
• Add dedicated SearchWorker goroutine so background data reloads don't block behind tree rebuilds