Handle shrank automation patterns correctly

[PhysSong]

Fix automation processing for shrank patterns. Retain the behavior before #3382 for those patterns.
Fixes #3800.

[lukas-w]

Would it make sense to move the check inside AutomationPattern::valueAt(...) so the fix is available to all callers?

[PhysSong]

@lukas-w I've tried something similar. It needs additional checks to keep current behavior in irrelevant places(ex. automation editor, note detuning). So I just chose this way for simplicity.

[lukas-w]

@PhysSong What's the issue with e.g. note detuning? Is the pattern's length not set correctly?

[PhysSong]

What's the issue with e.g. note detuning? Is the pattern's length not set correctly?

Yes. I detuned a two-bar note in piano-roll and the automation pattern have points over one measure. However, Its length is 192 ticks long(e.g. one bar long). Here's the project file:
detuning.mmp.zip

[lukas-w]

I see, thanks for explaining. I'd say we merge this and do further changes (if any) on master.

[lukas-w]

Merged and added unit tests for the bug via d146308.

[PhysSong]

@lukas-w In d146308, passing double caused linkage issue with Qt4 build. Fixed via 6cc118c.

[lukas-w]

Thanks @PhysSong. It's strange that the error only appears on macOS…

[PhysSong]

Unit test is disabled in Windows builds, and Linux Qt4 build is missing now. So only Mac + Qt4 fails.

[lukas-w]

Of course!