Add --archive-s3 <bucket> option

[Mr0grog]

This is pretty ugly, but it's a working first cut at the problem and fixes #663. This is suddenly more important since IA seems to be under a lot of load with peopole checking up on Trump-related website changes, and I'm seeing a lot more import-related failures on things that work fine on the processing side and fail to re-download in the DB importer. This will short-circuit that issue.

This ideally would go in a separate worker component, but we just don’t have the right data available from outside WaybackRecordsWorker right now. This would also ideally be tested more, but 🤷

[Mr0grog]

Turns out this is uploading data with an invalid binary/octet-stream media type. We should use the Content-Type header from the memento or, if not present, the sniffed media type (or we should just use the sniffed type?).

A quick look at Cloudpathlib suggests this might be hard. We could configure gzip encoding and a standard ACL for all uploads, but we’d have to construct a custom client for each upload now (I don’t think this is high overhead, but… I dunno?). Maybe need to drop down to boto3 and drop cloudpathlib (for this use case at least).

Also while I’m here, we should keep a list of seen hashes in memory so we don’t waste effort checking S3 if we encounter something a second time.