Move events to main thread and remove unneeded locks

[andydotxyz]

This means for race-free apps except for data binding and goroutines, which are covered later, separately.

Progresses #4654

Checklist:

• Tests included.
• Lint and formatter run with no errors.
• Tests all pass.

[Jacalz]

Nice work. Just one minor comment (not necessarily request for changes) inline. On more question: are there any parts where we should still support calling methods on different threads? Just asking so we don't get carried away and rip out too many locks at once :)

[Jacalz]

Should atomic types and things from the async package also be removed or is that perhaps better to have as a separate PR?

[andydotxyz]

I was wondering the same. I think later PR as it could impact semantics too

[Jacalz]

FYI: There seems to be compilation issues?

[andydotxyz]

FYI: There seems to be compilation issues?

Yeah I just noteiced I didn't compile the tests for last refactor. Updated

[andydotxyz]

On more question: are there any parts where we should still support calling methods on different threads?

I think it's just data binding which requires more work for various reasons.

[andydotxyz]

Looks like there may be a deadlock on Linux only - I'll check it out tomorrow

[Jacalz]

Seems to work fine on Linux for me. Maybe deadlock is for tests only?

[andydotxyz]

Seems to work fine on Linux for me. Maybe deadlock is for tests only?

Yes it was, or a codepath that was not executed on the testing. Fixed.

[coveralls]

coverage: 59.644% (-0.3%) from 59.964%
when pulling 017e085 on andydotxyz:fix/mergethreads
into 7fea2eb on fyne-io:develop.

[andydotxyz]

One thing led to another and I just refactored the event queue out in this PR too.

Fixes a few flakey tests :)

[Jacalz]

Nice. Well done. I love it when the diff of added vs removed looks like it does now. It also feels like the event queue removal made a huge performance difference compared to a previous version of this PR :)