[v4.y] Regenerate test TLS certs, stop skipping those tests

VCR was interfering with my attempts to connect to a real cluster,
despite `WebMock.enable_net_connect!`.
What's worse, it depended on order that tests executed.

[This won't be needed on master, we dropped the VCR test in ManageIQ#495.]

[v4.y] Ensure VCR doesn't leak into other tests

[v4.y] Test on Ruby 3.1

https://docs.github.com/en/packages/working-with-a-github-packages-registry/working-with-the-docker-registry

Followup to ManageIQ#540,
tests that it really fixes ManageIQ#525.

That bug only existed on master, not v4.y branch, but testing the behavior on both is good.

Yay, `docker run --priviledged` is allowed!  k0s starts up really fast too.
(only on linux, keeping regular `rake test` on macos because no `docker`.)

Sometimes minitest starts and then just hangs printing ...nothing?

    Run options: --seed 34017

    # Running:

    Error: The operation was canceled.

Github's default job timeout is a generous 6 hours!
https://docs.github.com/en/actions/learn-github-actions/usage-limits-billing-and-administration#usage-limits
But waiting is not productive and wasteful (and will burn our free minutes).

Setting outer job timeout longer for `bundle install`.
Hoping an inner SIGTERM received from while job is still running may allow
the test to print some traceback.  I have no idea where it's getting stuck.
(macos has no `timeout` command, but most or all stuck runs were linux.)

(I would suspect `exercise_watcher_with_timeout` but also saw same
symptoms on another PR without k0s tests)

Co-authored-by: Michael Grosser <michael@grosser.it>

shows stacktrace within Rake itself, not within test.

…er_ssl_verify