[WIP]: ddp pickle fix for learning rate finder

[SkafteNicki]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?
• If you made a notable change (that affects users), did you update the CHANGELOG?

What does this PR do?

Fixes #1831
This fixes the pickle error on ddp for the learning rate finder as described in the issue. I can get the learning rate finder to finish its search, however since the internal state is destroyed after ddp training so are the logged results. How can I get around this?

[williamFalcon]

i think it's time to add a shared memory queue before going into the process so we can bring stuff back out. this would also solve the test results issue.

Want to add that to this PR?

https://pytorch.org/docs/stable/notes/multiprocessing.html#reuse-buffers-passed-through-a-queue.

https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Queue

Probably create it in def .fit right before spawn?

[williamFalcon]

@SkafteNicki shall we merge this now and tackle the Q thing later?
or shall we push this fix to 0.7.7?

[SkafteNicki]

Push to 0.7.7 since the PR right now only is the solution to the pickle problem.

[mergify]

This pull request is now in conflict... :(