[ObjectStore] MultipartUpload::complete fails on WSL when writing to /mnt/

[madesroches]

Describe the bug
Using LocalFileSystem on WSL + BufWriter I get the following error:
file.metadata() fails after std::fs::rename with FileNotFound but is reported as an aborted upload

To Reproduce
Use object_store::buffered::BufWriter to write a big enough file to the windows filesystem when running though WSL.

Expected behavior
File write succeeds.

Additional context
Using object_store-0.11.0.
Our services are actually writing to S3, but the local dev environment writes to the local filesystem. I could not reproduce the bug when writing to the linux filesystem, but it happens every time I write a 50MB+ file to the windows file system with the exe running on the linux side.

[tustvold]

Possibly related to apache/arrow-rs#5094

My guess is that the WSL bind mounts don't quite match standard filesystem semantics, much like blobfuse, and so issues like this arise. It may be that a similar workaround to apache/arrow-rs#5094 is possible for multipart uploads. An interesting data point might be to confirm that this always occurs with the multipart codepath, e.g. by using ObjectStore::put_multipart manually.