Hold write lock during fs remove in gc

[jgrund]

This change holds the segement write lock while we remove stale files from the fs. This resolves a race condition where reads can try to reference files that no longer exist on disk.

Fixes #38

[marvin-j97]

I don't think this actually solves the underlying issue, where you can hold a value handle, then the blob file gets rewritten, then you resolve the value handle.

Also the read lock is only held shortly, so it could still read the file while not being locked.

[marvin-j97]

FWIW if the read lock was extended, it would probably solve the race condition here in value-log, but lsm-tree can probably still crash under certain circumstances (https://github.com/fjall-rs/lsm-tree/blob/main/src%2Fblob_tree%2Fmod.rs#L47). Basically you when you retrieve a value handle, you get locked out by GC and suddenly the blob file is gone.

[jgrund]

Thanks, I wasn't sure if I got the whole solution or not.