When rendering content for diffs backed by hosting services, we ended up

needing access to the parent DiffSet, its Repository, and that

repository's HostingServiceAccount. We ended up doing

HostingServiceAccount queries far too often, since we couldn't reuse

any fetched state.
To address this, we're now re-attaching the repository on each DiffSet

we load, so long as it's the same repository ID. We also attach the

DiffSet back to each diff comment, completing that chain. We had

processing code like this for file attachments and screenshots, so we

now have similar code for diff comments.
To reduce overhead, the per-comment processing code has moved into

local functions, which can just be provided to our prepared data for the

comment processing loop. This avoids an isinstance check per-comment.
While there, work was done to limit how many comments we have to

process. We no longer check screenshot comments if we don't have

screenshots, file attachment comments if we don't have file attachments,

or diff comments if we don't have a repository.
And finally, some code operating on DiffSet IDs has been simplified to

reuse existing state in one place and to avoid an unnecessary query in

another.

Unit tests pass.
Checked queries using Django Debug Toolbar. Verified that these

redundant queries were eliminated.