M2.1: Batch WAL writes

[SantanuKar43]

In fsync ALWAYS mode, instead of writing to wal file and fsyncing on each command, batch multiple requests and fsync them together.

In ALWAYS mode, batching emerges naturally from Go’s synchronization: each request acquires w.mu, appends its entry to the shared asyncBuffer, updates UncommittedLSN, and releases the lock. So multiple goroutines quickly accumulate writes into the same buffer without blocking on disk I/O.

The first goroutine that notices no flush is in progress flips isFlushing and signals the flusher via flushChan. The flusher then acquires w.mu (mutually excluding new writers), drains the entire buffer in one go, updates CommittedLSN while holding the sync.Cond lock (CommitSignal.L), and finally calls Broadcast() to wake all waiting goroutines.

Each waiting request, which is blocked on the condition variable until its LSN ≤ CommittedLSN, proceeds only after the batch is durably written. So batching is achieved by many goroutines serializing appends via a mutex, but amortizing a single disk flush across all of them, coordinated through a condition variable for durability acknowledgment.