Review and improve locks in Storage

[ikopylov]

RWLock on safe does protect active_blob + blobs updates inside, so the write lock should be acquired only when these 2 fields updating. But if we look at write_with_optional_meta function, it is incorrectly acquire write lock (https://github.com/qoollo/pearl/blob/master/src/storage/core.rs#L313). It should acquire read lock, because it does not update neither active_blob nor blobs fields.
Same thing here: https://github.com/qoollo/pearl/blob/master/src/storage/core.rs#L890 and here: https://github.com/qoollo/pearl/blob/master/src/storage/core.rs#L912. Probably, there are more places where write lock taken incorrectly.

Also, it is strange that blobs protected with additional RWLock. In every places this lock acquired right after acquire of RWLock on safe, thus can be removed.

This incorrect locking mechanism results in significant performance degradation

[idruzhitskiy]

RwLock on blobs can be removed. But lock can't be made read in write_with_optional_meta, since blob.write(...) requires mutable reference (due to index modification). But mutex on current_offset can be removed instead.

[idruzhitskiy]

Also any redundant write acquisition in most cases should be reported by compiler as redundant mut.

[ikopylov]

Exclusive lock should be as close to actual write to file as possible, so it is better to keep lock on current_offset and somehow avoid write lock in write_with_optional_meta