Reconsider the use of unsafe/safe for the FS layer

Ideally, we'd optimize for many concurrent reads, sacrificing concurrency for writes (which typically will just be a copy into a system OS buffer anyway and therefore very fast).

Start by locating the FS layer implementation and reviewing how unsafe and safe operations are currently used. Compare the existing read and write synchronization behavior with the issue's concurrency goal; done requires an agreed design and corresponding implementation changes, with validation of concurrent reads and writes.

Written by the indexing model from the issue text.