Amend RFC #36 with a concrete concurrency model

[whitequark]

Rendered

[whitequark]

Notwithstanding that there are no dependencies between any of these items, I have implemented most of the proposed concurrency model in amaranth-lang/amaranth#1231 and confirmed that it made it easier to reason about testbench concurrency.

[whitequark]

Note: I pushed some last-minute changes to until/repeat which were queued but not previously published (sorry about that!). In short, these methods now always advance the timeline. This is because it does not take away any expressive power (a single sim.get before the method doesn't add significant overhead even with an optimized implementation offloading the waiting to C++) but in turn reduces the likelihood of observing a glitch, no matter the context. Also, the stream use case is actually better served by the new API.

It also makes some intuitive sense that sim.tick().until(...) will perform strictly no less ticks than a bare sim.tick(), and seems like it'd be surprising otherwise.

[whitequark]

We have discussed this RFC on the 2024-04-08 core subsystem meeting. The disposition was to merge.