Fix mono delay

[Finomnis]

Fixes #844.

Solution

• Compensate for timer uncertainty by always waiting for one extra timer tick.
• Replace all .micros() with .micros_at_least() and all .millis() with .millis_at_least().

Open questions / Problems

• This is technically a breaking change for rtic-time, and indirectly therefore also for rtic-monotonics and rtic itself. This would require all of those to be updated to a new major version.
  • Answer: It's a soundness issue, so it is fine to release it as a patch.
• The previous versions technically did not comply with embedded-hal's definitions of DelayX. So should we yank those?
  • Answer: Yes. Let's yank previous versions.

[burrbull]

https://docs.rs/fugit/latest/fugit/trait.ExtU64Ceil.html

[Finomnis]

https://docs.rs/fugit/latest/fugit/trait.ExtU64Ceil.html

That doesn't fix the problem.

• 1ms tick rate, clock is at 0.99ms
• "900.micros_at_least()" -> 1ms delay scheduled
• actual wait time 0.01ms, although at_least() was used

To fulfill the requirement, both is actually needed.

With our fix, but without at_least():

• 1ms tick rate, clock is at 0.99ms
• "1200.micros()" - rounds to 1ms delay
• still only 0.01ms waited, although even with rounding, everyone would expect at least 1ms

[Finomnis]

Delays for blocking eh is still broken.

• Implement both delay_us and delay_ms and add one timer tick to the wait.

[burrbull]

Add more tests, please. Those cases which this PR fixes.

[Finomnis]

Add more tests, please. Those cases which this PR fixes.

I can add some to rtic-time. I can't really add some to the monotonic impls, as this would require an external time measurement separate from the microcontroller.

Would that be good enough then?

[burrbull]

Yes, please.

[Finomnis]

@korken89 still working on more tests. Need to rewrite the test to do a sub-tick test

[korken89]

Alright!

[Finomnis]

@burrbull I wasn't able to add tests to the blocking functions; those are hard to test reliably. But I now test all async functions. Is that enough for now?

EDIT: I added some tests for the blocking functions and they do work, but they are unreliable. So I disabled them by default.

[korken89]

Thanks for the effort!