Mark PR reviews as stale at push and allow to dismiss stale approvals

[davidsvantesson]

Fix #5997.

• If a push causes the patch/diff of a PR towards target branch to change, all existing reviews for the PR will be set and shown as stale.
• New branch protection option to dismiss stale approvals are added.

To show that a review is not based on the latest PR changes, an hourglass is shown:

I think there is a point in indicating reviews as being old/stale even if the branch protection does not have the setting to discard stale approvals.
Clean merges should not cause reviews to become stale, as they will not change the diff content of the PR. I have not found any smarter way than generate the diff before and after the push and compare (ignoring some lines). I tested it a bit (and compared to when GitHub marks stale reviews) and it seems to work.

New branch protection option:

[codecov-io]

Codecov Report

Merging #9532 into master will decrease coverage by 0.07%.
The diff coverage is 4.92%.

@@            Coverage Diff            @@
##           master   #9532      +/-   ##
=========================================
- Coverage   42.17%   42.1%   -0.08%     
=========================================
  Files         583     584       +1     
  Lines       77360   77485     +125     
=========================================
- Hits        32627   32623       -4     
- Misses      40721   40850     +129     
  Partials     4012    4012

Impacted Files	Coverage Δ
models/migrations/migrations.go	1.3% <ø> (ø)	⬆️
modules/auth/repo_form.go	43.47% <ø> (ø)	⬆️
modules/git/repo_compare.go	69.56% <0%> (-3.17%)	⬇️
services/pull/review.go	0% <0%> (ø)	⬆️
models/branches.go	45.5% <0%> (-0.39%)	⬇️
models/migrations/v118.go	0% <0%> (ø)
routers/repo/pull_review.go	0% <0%> (ø)	⬆️
routers/repo/pull.go	30.75% <0%> (ø)	⬆️
services/pull/pull.go	21.67% <1.36%> (-8.02%)	⬇️
modules/repofiles/update.go	41.28% <100%> (ø)	⬆️
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5b2d933...3d11d3a. Read the comment docs.

[lafriks]

We should save also commit sha on what PR was made so if old commit is reverted back we could remove stale flag

[davidsvantesson]

So the stale flag should be removed if there is a force-push to old commit, or if the old commit is reverted in a new commit, or maybe both? I tested on Github and I don't think they handle that. When a new commit is made Github adds an event:

userA dismissed userB’s stale review via 0123abc 2 minutes ago
and after that a new review is always required. But of course we could try to do better than GH.

Would this mean at every new push all old reviews (latest of each reviewer) has to be checked again if they are valid?

[lafriks]

That could be improved in other PRS, more important and quite easy to impelemt that we would set to stale or not based if review commit sha equals currently latest (pushed)

[lafriks]

Later we could improve so that we could check to set stale only if diff actually has changed etc but would be faster to merge this in smaller steps

[davidsvantesson]

The additional code needed to add dismiss stale approvals to branch protection options was small, so I decided to add it in this PR as well.

[6543]

ping lgtm bot