Remove content app redirect for manifest fetching

[gerrod3]

Fixes #1974
https://redhat.atlassian.net/browse/PULP-486

To properly remove the redirect for manifest fetching I had to remove the PullThroughCaching redirect logic as well. Now the entirety of PullThroughCaching lives on the registry_api side, no more content app logic needed. These are the three scenarios we were handling for pull-through caching:

1. Manifest tag fetch:
   Upon first fetch you need to create the tag and download the manifest and initiate a task to download the rest of the image in the background (tag=None, local_manifest=None). If the upstream manifest already exists on Pulp you simply need to create the tag and add it and the image's content to the repo (tag=None, local_manifest is not None). On the second fetch of a tag with pull-through you need to check to see if the tag has been updated on the remote. There are three scenarios for this: one - new upstream manifest that needs to be downloaded to Pulp (tag is not None, local_manifest=None), two - new upstream manifest already in Pulp needs to be added to repo (tag is not None, local_manifest is not None), and three - no changes on upstream (tag is not None, tag.manifest.digest == local_manifest.digest). If the tag doesn't exist on the remote, but we have the tag locally then we should serve our local copy (tag is not None, fetch_manifest raised ApiException).
2. Manifest digest fetch:
   When fetching a manifest by digest we either have it or we don't. If we don't, we check the upstream if it's there. If yes, then we check if the manifest is in Pulp locally and just needs to be added to the repo, or if we need to download the whole image in the background.
3. Blob fetch:
   When fetching a blob and we don't have it in the repository we check to see if the remote has the blob. If so, we create the blob as an on-demand content and redirect to the content-app for it to finish streaming the blob from the upstream. We also add it to the pending_blobs to speed up future requests.

📜 Checklist

• Commits are cleanly separated with meaningful messages (simple features and bug fixes should be squashed to one commit)
• A changelog entry or entries has been added for any significant changes
• Follows the Pulp policy on AI Usage
• (For new features) - User documentation and test coverage has been added

See: Pull Request Walkthrough

[dralley]

can you explain (and maybe document) this a bit better? What is EMPTY_BLOB and why does it redirect?

[gerrod3]

It's a layer that has nothing in it. Each layer(blob) is a gzip-compressed tar and the empty blob is a no-op layer used in images, so we handle it specially and have the empty compressed bytes ready to serve if requested. See the _empty_blob() method in registry.py, there are still bytes that need to be served so that is why it was originally redirected to the content app.

[dralley]

Missing doc comments - would be nice if all functions added here had them