This PR fixes a lock/unlock mismatch in BatchWithFlusher.Set and Delete that can cause a runtime panic when Write() returns an error on the flush path.

Both methods use defer b.mtx.Unlock() while also manually unlocking the mutex prior to calling Write(). If Write() fails, the function returns while the mutex is already unlocked, and the deferred Unlock() triggers:

fatal error: sync: unlock of unlocked mutex

Changes

• Reworked the flush path to always re-acquire the mutex before returning.
• Ensures that all return paths maintain the invariant expected by the deferred Unlock().

Impact

• Converts a hard panic into normal error propagation.
• Preserves existing behavior on successful writes.
• Makes BatchWithFlusher safe when Write() can legitimately fail (e.g., alternative backends, injected faults, or stricter error handling).

Notes

This change intentionally does not alter higher-level error handling semantics for failed writes; it only ensures mutex correctness and stability.

If you want, I can also:

• add a minimal unit test demonstrating the original panic,
• tune the tone for Cosmos governance style,
• or help you anticipate maintainer questions (e.g. about panic-safety or write-failure semantics).