fix(fsspec): _info() should honor self.dircache

[fvaleye]

Summary

This PR eliminates unnecessary network calls in FsspecStore._info() by ensuring it consults the directory cache before falling back to a network request, bringing it into alignment with standard fsspec implementations.

Currently, FsspecStore._info() issues an unconditional head_async() and ignores self.dircache, even though AbstractFileSystem._ls_from_cache is documented as the standard way for callers to surface pre-known file metadata.

For comparison, both s3fs and gcsfs consult _ls_from_cache() first and only fall back to a network call on miss. obstore.fsspec.FsspecStore is the outlier here: pre-populating fs.dircache (or having it populated by a prior _ls) is ignored.

Notes

When the parent directory is cached, and the requested child is not in the listing, _info() now raises FileNotFoundError immediately instead of falling through to a _ls(). This matches s3fs._info and the documented contract of _ls_from_cache: a cached parent listing is authoritative, so a missing child definitively does not exist.

[kylebarron]

Thank you!

This is why fsspec is so awful to work with. Each implementation needs to know specifics of other internals of fsspec to be correct. Even if fsspec decides to implement caching, it should do it at a level that doesn't require implementors to know about it!

(FWIW I'm happy to merge PRs fixing fsspec handling, but in general I recommend using the non-fsspec API)