feat: Remove object store uses from Wren

[bjchambers]

Summary

Currently, Wren uses object stores for a variety of things. At least one of which is downloading a file locally prior to getting the schema via FileService. As of #479 Sparrow is able to get the schema for any URL, including s3:// and gs://, etc. via object_store. And doing so will only need to retrieve the footer of Parquet files.

We should at least switch Wren to not downloading files locally for this use case and also consider removing all object store dependencies from Wren, so that Sparrow is the only thing with object store credentials, etc.

[epinzur]

Wren does not download files locally for the FileService.

Currently wren only copies a file into the fileSystem owned by Wren (local or objectStore) on Table Load. And the default behavior for this will be changed soon to leave the file in-place.

[epinzur]

wren also uses object store to:

• get file identifiers to verify file uniqueness (unique ids from s3, gs, or md5 when working locally)
• pre-sign query responses when output is written to object storage
• check for file existence before sending a file-service request

[bjchambers]

Ah. Great. So if I understand correctly, Wren uses object stores for:

1. Signing output URLs before returning them, so the user can read them without credentials.
2. Copying files into the "owned" object store on table load, if necessary.

With your change, 2 is no longer the default. And the first one could be (conditionally) disabled in cases where the user already has credentials for the file store (eg., doesn't need a signed URL).

So it sounds like no work is needed here to avoid copying files to disk (beyond #472). We can keep this issue for tracking the remaining object store dependencies within Wren, but that can be handled later. Correct?

[epinzur]

That is correct. We can delay this work into the future.