Refactor the put_object method to avoid temporary file write

[brunocalza]

While uploading an object, we first write all contents in a temporary file before sending it to Hoku.

This is clearly bad for performance, especially for big files. Ideally, we would stream the bytes from the request directly to Hoku. We didn't do this in the first version because of:

• The Hoku SDK method add_reader required a AsyncSeek (this is not true anymore)
• We need a way to calculate the md5 of the file and pass it as metadata

So, writing to a temp file was the easiest way out of this.

One solution is to wrap the body request in an AsyncRead implementation that calculates the MD5 and passes it to add_reader. But we wouldn't be able to get the final md5 hash value before calling add_reader. So in addition to that, we would need a way to add metadata to an existing Hoku object (we don't have that Hoku operation right now).

[stbrody]

Most of this work would live in the S3 adapter and so could be done after audit, but to do this we would need a feature to update the metadata of a blob after it is uploaded, which lives onchain and would need to happen before feature freeze.

@brunocalza can you file a separate ticket for the work to update metadata and put it in Triage so we can consider if we want to get it in for audit?

[brunocalza]

@brunocalza can you file a separate ticket for the work to update metadata and put it in Triage so we can consider if we want to get it in for audit?

@stbrody recallnet/ipc#373

[stbrody]

The actor changes were indeed done in time for audit, so we can consider doing the changes to basin-s3 post-audit.