Having a distinct class just for that was a bit overkill

Rename classes to (hopefully) clearer names while we're doing that.

Closes gitaly#524

For some old merge requests, we don't have enough information to figure out the
old blob and the new blob for the file. This means that we can't highlight the
diff correctly, but we can still display it without highlighting.

We didn't have a fallback case before, because we believed the conditions were
exhaustive. They weren't, so we can always fallback to not previewing.

* Added keyword arguments to truncated_diff_lines method to allow for using highlighting or not (html templates vs. text)
* Tweaked templates for consistency and format appropriateness

Previously the `truncated_diff_lines` method for outputting a discussion diff took in already highlighted lines, which meant it wasn't reuseable for truncating ANY lines. In the way it was used, it also meant that for any email truncation, the whole diff was being highlighted before being truncated, meaning wasted time highlighting lines that wouldn't even be used (granted, they were being memoized, so perhaps this wasn't that great of an issue). I refactored truncation away from highlighting, in order to truncate formatted diffs for text templates in email, using `>`s to designate each line, but otherwise retaining the parsing already done to create `diff_lines`.

Additionally, while notes on merge requests or commits had already been tested, there was no existing test for notes on a diff on an MR or commit. Added mailer tests for such, and a unit test for truncating diff lines.

Steps to reproduce:

1. Start with a repo with an image
2. Add a commit to delete the image
3. Add another commit to replace the image with another image

In a diff comparison, we really just compare about what the image was before the diff, not
the direct parent of the last commit. This MR fixes that.

Closes #3893, gitlab-org/gitlab-ee#678

Signed-off-by: Rémy Coutable <remy@rymai.me>

In this case comparing old_path and new_path is not enough because there
are two entires that match.

In this specific case using file_path as a cache key is not enough,
because there are two entries with the same path.
Closes #21610.

Instead calling diff_collection.count use diff_collection.size which is cache on the diff_collection

Introducing the concept of SafeDiffs which relates 
diffs with UI highlighting.

When rendering a list of diff files, skip those where the diff is over
10 KB and provide an endpoint to render individually instead.

Extract parts of LegacyDiffNote into DiffOnNote concern and move part of responsibility to other classes

This builds on the changes introduced in
https://gitlab.com/gitlab-org/gitlab_git/merge_requests/72 and results
in merge requests with large diffs (e.g. due to them containing minified
CSS) loading much faster.

* Use commit objects instead of IDs when generating diffs
* Use proper references when generating MR's source and target
* Update broken specs

Now we apply syntax highlighting to the whole old and new files.
This basically help us to highlight adequately multiline content.

* It is more performant given now we process all the diff file instead
  of processing line by line.
* Multiline comments are highlighted correctly.