Activate nightly workflow for random and stress tests

[kdeldycke]

In #3139 we added stress tests and in #3151 we added randomization and parallel tests.

We decided to not activate them by default in #3151 (comment) and rely on contributors diligence to run them by hand to catch regressions. In practice nobody does. I even often forget to run them on my own during my sessions, only focused on the task at hand.

I'd like to re-assess this decision in light of regressions that could have been caught if these test were active. See the following PRs:

• fix I/O operation on closed file error with CliRunner and echo_via_pager #3482
  Regression introduced in 8.4.0, could have been caught by tests/test_stream_lifecycle.py.
• Fix pager test flackiness and add stress tests for echo_via_pager #3499
  New stress test covering the root cause of fix: Skip flaky pager test on macOS with free-threaded Python 3.14t #3470.
• fix: Skip flaky pager test on macOS with free-threaded Python 3.14t #3470
  The bug exists because CI doesn't run tests in parallel.
• fix a race in click.testing.StreamMixer's finalization, seen in a multi-threaded context #2991
  The reason why stress test exist: I originally only hit the race about once in 10k tries, so this seems difficult to test systematically..
• NamedTextIOWrapper: stop closing buffer #3139
  Fix the regression from fix a race in click.testing.StreamMixer's finalization, seen in a multi-threaded context #2991, where only with the introduction of stress test keeps from recurring.
• Fix pager test leaks #3238
  The random test already caught a leak nobody noticed via the default suite.

Also why even support parallel tests if we don't exercice them regularly? (ref: #2899)

Currently these tests adds ~15 minutes run-time. If you find this too long and invasive, I propose a compromise and run them nightly. But my preference would be to run them as part of the standard CI to incentivize contributors. And reduce surprises and and follow-up fixes.

Given the reversal in policy this change imply, I'm aiming for the next 8.5.0 minor release. I'm opening this PR as a way to discuss the best strategy forward.

[Rowlando13]

@kdeldycke I agree with the nightly run.

[kdeldycke]

I think we were overly cautious just running these tests nightly as in the end they only take 5 minutes, not 15: https://github.com/pallets/click/actions/runs/26496766042/usage . To me the cost-benefit is excellent to trigger on every PR: a 5 minute wait is nothing in comparison to the effort and synchronization time required to fix leaky issues from non-core maintainers. But I will let that decision for later.

Thanks @Rowlando13 for the merge!