Optimize `jj log` generation via async execution and caching

This PR significantly improves the performance of jj-log.nvim by reducing synchronous blocking operations and optimizing how often the jj log subprocess is spawned. ## Key Optimizations * Replaced synchronous file I/O checks: Swapped the blocking vim.fn.filereadable() with a quick check on the buffer type (vim.bo.buftype). The plugin now naturally ignores unbacked buffers (terminals, help files, etc.) without hitting the disk. * Directory Caching: Refactored .jj repository detection into a new helpers.lua module. The upward directory traversal now utilizes non-blocking vim.uv.fs_stat and caches all intermediate directories. * Subprocess Debouncing: Implemented a per-repository lock to prevent the plugin from spawning overlapping jj log shell commands if multiple autocommands fire concurrently. * Command Output Caching: Added a description cache (cache_timeout_ms, defaulting to 2 seconds). Subsequent buffer events within this window will immediately return the cached string rather than spinning up a new shell process. * Synchronized Split/Tab Updates: When the async jj log command finishes, the plugin now updates vim.b.jj_desc for all valid open buffers belonging to that repository simultaneously, ensuring status lines stay perfectly in sync. * Graceful Failures: Wrapped Neovim API calls in pcall() so the plugin fails safely if a buffer is closed before the async shell command completes.