When there is an Evergreen failure on a test that uses tiered storage, we should capture any WT files in object storage (e.g., S3) as part of the test artifacts so that we can refer to them during debugging.

I can imagine several ways of doing this – each with benefits and challenges.

1. After test failure we find and copy cloud objects to a directory that is included in the artifacts tarball
2. We don't delete the cloud objects, and instead save something with the artifacts that would tell an engineer where to find them.
3. We implement a shim layer that saves a copy of any object when it is written to the cloud – or alternatively that save the original local file after it has been uploaded instead of deleting it.

WT-10786 is an example of a failure where it would have been useful to have the access to a cloud object that (apparently) generated a checksum mismatch on read.