This change rearranges some methods and adds new ones to the GitHub client

and hosting service to improve the way we fetch and build GitHub changes.

The logic for all of this used to live in GitHub.get_change(), and now

we've pulled it out into general helper methods and improved the way

we build and process GitHub diffs.
We have a new GitHubClient.get_diff_content() method

that handles getting the diff content for a change and expanding the

blob SHAs in it. get_change() used to build the diff manually by

looping over the files in the change, but now we fetch the raw diff

content directly from GitHub's API and replace the index lines with

full SHA index lines, in pretty much the same way that we do for

Forgejo.
Our old code also didn't handle commits without a parent, aka the root

commit of a repo, but now we handle this and allow for review requests

to be created from these commits. And files without textual changes,

like binary files, mode-only changes, and pure renames are now included

in the diff, whereas before they were dropped. We still don't include

the actual content of binary file changes, but we could add that later.
We also now cache blob SHAs onto the client and in the cache backend,

to help reduce the amount of API requests we make for file

existence checks and diff processing. This will be important for pull

request integration, where we're likely to fetch a lot of commits and

diffs at once.
The get_change() unit test was updated to use a raw diff from GitHub

instead of the JSON from before, and checks for a binary file and

a rename header.

Ran unit tests.
Used for GitHub pull request integration.
Tested creating a new review request from a GitHub commit through

  the UI (a root commit, and a commit that has binary file and .py

  file changes).