This adds a method to track errors that can be recovered from in
sentry.

It is useful when debugging performance issues, or exceptions that are
hard to reproduce.

These can be a `BatchLoader` which is proxying a nil, while not being concrete
nils themselves.

Previously, we kept them all in the cache. We don't need the highlight results
for older diffs - if someone does view that (which is rare), we can do the
highlighting on the fly.

Rugged sometimes chops a context line in between bytes, resulting in the context
line having an invalid encoding: https://github.com/libgit2/rugged/issues/716

When that happens, we will try to detect the encoding for the diff, and
sometimes we'll get it wrong. If that difference in encoding results in a
difference in string lengths between the diff and the underlying blobs, we'd
fail to highlight any inline diffs, and return a 500 status to the user.

As we're using the underlying blobs, the encoding is 'correct' anyway, so if the
string range is invalid, we can just discard the inline diff highlighting. We
still report to Sentry to ensure that we can investigate further in future.

Due to the refactoring in !16082, `Blob#batch` no longer falls back
to a default `blob_size_limit`. Since `Repository#batch_blobs` was using
a default `nil` value, this would cause issues in the `Blob#find_by_rugged`
method.

This fix here is to be consistent and use a non-nil default value in
`Repository#batch_blobs`.

The problem was masked in development and tests because Gitaly is always
enabled by default for all features.

Closes #41735

Old merge requests can have diffs without corresponding blobs. (This also may be
possible for commit diffs in corrupt repositories.)

We can't use the `&.` operator on the blobs, because the blob objects are never
nil, but `BatchLoader` instances that delegate to `Blob`. We can't use
`Object#try`, because `Blob` doesn't inherit from `Object`.

`BatchLoader` provides a `__sync` method that returns the delegated object, but
using `itself` also works because it's forwarded, and will work for
non-`BatchLoader` instances too. So the simplest solution is to just use that
with the `&.` operator.

- handle unchanged empty lines in inline diff

If a merge request was created with a branch name that also matched a tag name,
we'd generate a comparison to or from the tag respectively, rather than the
branch. Merging would still use the branch, of course.

To avoid this, ensure that when we get the branch heads, we prepend the
reference prefix for branches, which will ensure that we generate the correct
comparison.

After installing a new gem, batch-loader, a construct can be used to
queue data to be fetched in bulk. The gem was also introduced in both
gitlab-org/gitlab-ce!14680 and gitlab-org/gitlab-ce!14846, but those mrs
are not merged yet.

For the generation of diffs, both the old blob and the new blob need to
be loaded. This for every file in the diff, too. Now we collect all
these so we do 1 fetch. Three `.allow_n_plus_1_calls` have been removed,
which I expect to be valid, but this needs to be confirmed by a full CI
run.

Possibly closes:
- https://gitlab.com/gitlab-org/gitlab-ce/issues/37445
- https://gitlab.com/gitlab-org/gitlab-ce/issues/37599
- https://gitlab.com/gitlab-org/gitlab-ce/issues/37431

Having a distinct class just for that was a bit overkill

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

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.

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

Remove superfluous lib: true, type: redis, service: true, models: true, services: true, no_db: true, api: true

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

Use the EnforcedStyleForMultiline: no_comma option.

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

This reverts commit 96bef54154e669f9a3e92c3a4bc76c0be3a52e48.

Reason for renaming is to comply with naming convention of services in
codebase.

Backport changes from the EE-only squash implementation, which would
otherwise conflict when merge CE into EE.

<https://gitlab.com/gitlab-org/gitlab-ee/merge_requests/1024>