`from_thread.run(_sync)?` fails in any thread running Trio

hi! from_thread.run(_sync)? uses a heuristic to prevent the user from accidentally deadlocking their event loop by writing code like https://github.com/python-trio/trio/pull/1574/files#r435039316.

the heuristic is currently much more aggressive than is required to prevent this deadlock, however. it has the side-effect[^1] of preventing one from calling from_thread.run(_sync)? in any thread running trio, even though such calls are almost always safe and deadlock-free.

[^1]: citing the RuntimeErrors reported in from_thread.run(_sync)?'s docstrings, this would appear to be an unintended side-effect of the current heuristic.

it would be useful to be able to call from_thread.run(_sync)? from trio threads[^2], but the current heuristic blindly prevents this. i'd like to propose relaxing from_thread.run(_sync)? to allow calling from_thread.run(_sync)? from trio threads. (one use-case of this is https://github.com/agronholm/anyio/issues/525.)

[^2]: for example in order to call Nursery.start_soon in a thread-safe manner.

Start with the from_thread.run(_sync)? implementation and its docstrings, then review the heuristic that raises RuntimeError when called from Trio threads. Check how the proposed relaxation affects calls such as Nursery.start_soon and verify that the documented deadlock prevention still holds. Done means safe calls from Trio threads are allowed without weakening the intended protection.