perf: Use Memory-Mapped Files (mmap) for Zero-Copy Hashing & Fast Scanning

[kavix]

🎯 Goal

Replace standard os.ReadFile / io.ReadAll with memory-mapped file access (mmap) during file hashing and diff generation for files larger than 1MB.

💡 Why

Standard file reading allocates Go heap memory buffers and copies bytes from OS kernel space to user space.
With mmap: the file is mapped directly into process memory space, allowing the CPU to hash bytes directly from OS page cache with zero heap allocations and zero memory copies.

🛠️ Implementation Steps

1. Add golang.org/x/sys/unix / golang.org/x/exp/mmap wrapper in internal/util/mmap.go.
2. Use mmap in internal/objects/store.go and internal/api/diff.go for files exceeding 1MB.
3. Ensure proper memory unmapping (munmap) on deferred cleanup.
4. Benchmark heap allocations (allocs/op) and CPU cycles during snapshot saves.