Since we build again after instrumentation

- using $GITHUB_RUN_ID-$GITHUB_RUN_ATTEMPT to keep each re-run unique
- Previously, cypress used to just return "Run finished" (without actually re-running the test) after re-running the GA build (without updating the code). This used to give false impression that all tests are passed.

since docs are moved to wiki

once a day some job gets stuck and default timeout is 6 hours.

Changed timeout to 1 hour which is 3-4x more than max running time of
all jobs.

This reverts commit b9ebe88c5a.

* ci: fail CI if asset bundling fails

* chore: formatting

Co-authored-by: Suraj Shetty <13928957+surajshetty3416@users.noreply.github.com>

minor changes:
1. quitely checkout new tags
2. avoid changing directory
3. use pip instead of bench to install deps.

In the patch test, instead of migrating to the current SHA directly.
Migrate to the latest release (with code at the time of release) and
then migrate to last commit in PR.

Why? When directly migrating, the previous patches that would've run
with old code have access to the new code and it's not realistic. This
change brings it a little bit closer to reality. Hence missing failures
that would occur with IRL installations.

* Bumping iPython by a minor version broke 3.6 installs for us via https://github.com/frappe/frappe/pull/14192
* We could just add another line in requirements.txt to solve this, but
  since PY36 is reaching end of life by end of this year and release of
  3.10 is just around the corner, might as well just drop it now than
    later
* Frappe v14 would probably have the support range of 3.7-3.10/11 given
  when we release it. Maintaining dependencies for such a large range
  can become cumbersome

GitHub action's Ubuntu 20.04 image has mysql8 tooling which is not
compatible with mariadb.

Why?
- Coveralls had login issues and we were not able to view files to find the actual uncovered lines
- Codecov has better reporting (status checks & commit diff coverage, ), UX (better folder/file navigation, faster page loads), and has more features.
- Codecov will not require additional step of
coverage wrap-up for parallel builds (it manages it internally somehow).

Issue: No PR number is detected on Push events and all builds are
skipped. We want the opposite to be true, so we're running all builds on
merges in hopes of testing better.