refactor(threads): avoid unnecessary scheduler invocations

[elenaf9]

Description

This PR implements some changes and checks to avoid unnecessary scheduler invocations:

• the scheduler is now triggered in set_state as a central point
• if a thread changes into ThreadState::Running, only trigger the scheduler if the thread has a higher priority than the currently running one (new)
• always trigger the scheduler when a thread changes from ThreadState::Running into another state because this only ever happens if a currently running thread gets blocked
• only trigger the scheduler in yield_now if the runqueue contains any other threads

Issues/PRs references

Extracted from #397.

Benchmarks

(ticks per iteration)
bench_sched_flags is PR'd in #456

nRF52840dk:

benchmark	main	this PR
bench_sched_yield with two threads	560	580
bench_sched_yield with one thread	235	98
bench_sched_flags	1848	1461

RPI-Pico

benchmark	main	this PR
bench_sched_yield with two threads	631	647
bench_sched_yield with one thread	243	78
bench_sched_flags	1739	1495

The additional checks come a some cost when the scheduler invocation is really needed, as visible in the bench_sched_yield benchmark.
But I think this is tolerable, because the actual context switching logic (i.e. sched) is not affected, and in many cases we save an unnecessary scheduler call that would be much more expensive.

Open Questions

See self review.

Change checklist

• I have cleaned up my commit history and squashed fixup commits.
• I have followed the Coding Conventions.
• I have performed a self-review of my own code.
• I have made corresponding changes to the documentation.

[elenaf9]

Rebased so that recently merged bench_sched_flags can be run on this branch.

[kaspar030]

LGTM!