push: skip already existing files in the s3 bucket on push

[mvo5]

[I leave this in draft for a bit, I'm using this code and will monitor its behavior. I tweaked it to use COPY-to-self so that the timestamp gets an update, I suspect that is the reason why we did the full upload before(?) - i.e. have time based expiration policies in S3]

When we push data (i.e. sha254-abcd) files to the s3 bucket we
currently do this unconditionally. But if the file is already in
the bucket this is unnecessary. So try to "touch" it first and
only upload if the file is not already there. This saves some
time even on a fast connection (and bandwidth of course).

Note that we use "touch" here because we want to make sure the
object metadata gets updated so that any time-based expiration
policies are still honored.

[achilleas-k]

There should be no case where a file with the same path/name needs to be updated since they're all content-addressable (IIRC), so this LGTM!

[mvo5]

Actually - one reason why this might be desired is that if a "older-than-n-days" policy to auto-delete objects is used. The constant rewrites ensures that they are "fresh" I suppose. Its just a guess but it might be a rason - if it is we should add a comment :) I tweaked the code now to s3c.copy_object() instead of head_object() which should fix this concern.

[achilleas-k]

FTR, we don't have a retention policy, but this is a good idea regardless.

[achilleas-k]

I know that copying an object onto itself is a common way to touch files in S3, but a comment here would be helpful for the future.

In the images CI cache, I also did a metadata key touched=<timestamp> because I think I read somewhere that without any metadata change, the actual modification time of the object wont change. Not sure if I'm remembering that correctly.