Improve mechanism to reset the seed after sanity check

[awaelchli]

What does this PR do?

An improvement to #7014
Fixes #11840

The previous approach was to reset the seed to the global seed value that was previously set with seed_everything.
https://github.com/PyTorchLightning/pytorch-lightning/blob/6f22b3623c28028026b3cb8bb534c1ebca9c5ac8/pytorch_lightning/trainer/trainer.py#L1375-L1377

However, this would lead to multiple subsequent fits to produce identical results. The expected behavior however is:

# Scenario 1
seed_everything(1)
trainer.fit()
trainer.fit()  # different results expected, since we didn't explicitly re-seed

# Scenario 2
seed_everything(2)
trainer.fit()
seed_everything(2)
trainer.fit()  # same results as first fit expected, since we explicitly re-seeded

This PR enables both scenarios to work as expected. An existing test was modified to enable scenario 2.
In addition, there is a test test_multiple_lr_find_calls_gives_same_results for the LR finder that was relying on the previous behavior. To preserve that, we now also properly reset the random state after tuning.

Does your PR introduce any breaking changes? If yes, please list them.

Theoretically yes. Anyone who relied on the previous behavior will get different results now.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

cc @Borda

[tchaton]

Super neat !

[tchaton]

LGTM !

[carmocca]

@awaelchli Pushed a commit fixing mypy. I think we can treat the rng keys as required.

@rohitgr7 mind reviewing?