Add BenchmarkCloneLargeRepo and BenchmarkCloneDeepRepo to measure cloneperformance when dealing with repositories containing thousands of files.These benchmarks exercise the resetWorktree codepath, which is currentlyprocessing files sequentially. This establishes a baseline for measuringthe performance improvement from parallelizing file checkout operations.With ~35,000 files in production repositories, the sequential checkoutbecomes a significant bottleneck. These benchmarks will help quantifythe improvement from parallel checkout implementation.

Make the Tree struct safe for concurrent access by multiple goroutines,addressing race conditions in lazy initialization of internal maps whilepreserving memory efficiency.Changes:1. Added sync.Once to guarantee single initialization of t.m map2. Changed t.t from map[string]*Tree to sync.Map for thread-safe cache3. Removed mutex from worktree.go parallel checkout (now unnecessary)Thread-safety guarantees:- entry() uses sync.Once for one-time t.m initialization- FindEntry() uses sync.Map for concurrent path cache access- Multiple goroutines can safely call FindEntry() simultaneouslyPerformance characteristics:- First FindEntry() call: builds map once (one-time cost)- Subsequent calls: lock-free reads from t.m (fast path)- sync.Map optimized for read-heavy workloads (typical usage)- Memory overhead: slight increase by using sync.Map instead of Map

Improve checkout performance for large repositories by parallelizing filecheckout operations. Each worker goroutine gets its own independent objectstorage instance, eliminating contention on packfile readers and enablingtrue concurrent processing.Implementation:- Added worker pool sized to runtime.GOMAXPROCS(0)- Created thread-safe createWorkerTree() called from within each goroutine- Each worker has independent packfile file descriptors and object cache- Made indexBuilder thread-safe with sync.Mutex for concurrent updatesKey design decisions:1. Per-worker isolation: Each goroutine creates its own filesystem.Storage   instance by calling createWorkerTree() directly. This provides:   - Separate file descriptors for packfile access   - Independent LRU object caches   - No shared mutable state between workers   - Eliminates packfile reader contention   - Cleaner encapsulation with storage created within goroutine scope2. Thread-safe createWorkerTree(): Designed to be safely called concurrently   from multiple goroutines. NewStorage() creates isolated instances, and   the shared billy.Filesystem is only used for read operations.3. Thread-safe indexBuilder: Protected Add() and Remove() methods with   mutex since multiple workers update the index concurrently.4. Graceful fallback: For non-filesystem storage (e.g., memory-backed),   falls back to shared tree without breaking functionality.The speedup comes from:- Parallel zlib decompression across CPU cores- Concurrent disk I/O for packfile reads and file writes- Better CPU utilization during checkout operationsThis change works in conjunction with the thread-safe Tree implementation(sync.Once + sync.Map) to enable safe concurrent access to git objects.Scalability:Performance improvement scales with CPU count. Repositories with morefiles and/or bigger files will see greater benefits. The approach handlesboth small repos (minimal overhead) and large repos (significant speedup)effectively.