Handling `breakpoint()` usage from sync code

[goodboy]

With the introduction of our multi-process debugger system in #126 via await tractor.breakpoint(), we probably need to consider how a user can call the standard breakpoint() from sync code in multiple (sub)actors.

There shouldn't be a conflict as long as multiple agents don't request the tty lock simultaneously at which point the sync breakpoint() may clobber some other actor's tty i/o.

An initial off-the-cuff idea is to just use trio.from_thread.run() from a new thread that is spawned at the point of a sync breakpoint() (obviously by overriding the default hook). We can also try just scheduling the parent tty locking sequence using Actor._service_n.start_soon() like we do normally in the root actor and it may work just fine?

The main conceivable issue with the latter will be of course that locking / queuing for the tty lock won't work with just the sync code but if we were to first launch a thread with trio.to_thread.run() which then calls the normal tractor.breakpoint(), blocking until the tty lock is release, it might just work?

Needs some experimentation for sure.

[goodboy]

Partially solved by #194 (in terms of avoiding issues with cancellation once inside pdb).

Relates heavily to #193.