Ensure that signed and unsigned S3 requests can coexist

From @zhangfengcdt in https://github.com/apache/sedona-db/pull/44 :

```
// TODO: Consider registering object stores per-bucket instead of per-scheme to avoid
// authentication conflicts. Currently, if a user first accesses a public S3 bucket with
// aws.skip_signature=true and then tries to access a private bucket, the cached object
// store will still have skip_signature enabled, preventing authentication to the private
// bucket. A per-bucket registration approach would solve this by using bucket-specific
// cache keys like "s3://bucket-name" instead of just "s3://".
```

(Opening an issue so that we can remember/discover this and circle back!)

Start with the object-store registration TODO quoted from pull request 44 and trace how the cached S3 store is reused across requests. Reproduce the sequence of opening a public bucket with aws.skip_signature=true and then a private bucket, and use that behavior to define completion: signed and unsigned S3 requests must coexist without authentication settings leaking between buckets.

Written by the indexing model from the issue text.