feat(services/s3): Add rename support for s3 service

[jihuayu]

I add rename support for the s3 service.
Because s3 doesn't have the native rename API, so we need to copy first and delete the old file.
I also added the rollback logic when we delete the old file failed.

[jihuayu]

It seems like CI fail in main. I will re-open this PR after the CI passes.

[Xuanwo]

Thanks for the PR. The capacity in backends is used to hint features that storage services support natively.

We don't need to implement rename in backend as we can do it at operator level.

For example: send rename if it's natively supported, otherwise fallback to copy and delete.

[jihuayu]

Thanks for the PR. The capacity in backends is used to hint features that storage services support natively.

We don't need to implement rename in backend as we can do it at operator level.

For example: send rename if it's natively supported, otherwise fallback to copy and delete.

OK, I will close this PR. Do we need to update the documentation to indicate that we do not support the rename operation for S3?

[Xuanwo]

Do we need to update the documentation to indicate that we do not support the rename operation for S3?

I feel like it's clear enough the s3 doesn't support rename.

OK, I will close this PR

Would you like to implement the simulate logic for rename in CompleteLayer?

opendal/core/src/layers/complete.rs

Lines 556 to 563 in 0f316b4

	async fn rename(&self, from: &str, to: &str, args: OpRename) -> Result<RpRename> {
	let capability = self.meta.full_capability();
	if !capability.rename {
	return Err(self.new_unsupported_error(Operation::Rename));
	}
	self.inner().rename(from, to, args).await
	}

[jihuayu]

Would you like to implement the simulate logic for rename in CompleteLayer?

I will do it.
Do we treat it as the default behavior? Or do we need to add a flag to indicate the use of simulation?

I feel like it's clear enough the s3 doesn't support rename.

In our doc, there is no strikethrough in the s3 rename capability. But some others service has a strikethrough.
I think this could lead to misunderstandings that we simply do not currently support S3 rename operations temporarily.
So, I think we need to add a strikethrough to the s3 rename capability.

[Xuanwo]

Do we treat it as the default behavior? Or do we need to add a flag to indicate the use of simulation?

We mark this different via full_capability and native_capability