in generators/coroutines, save away the current exception in the 'return' case of finally clauses as 'return' actually raises an (Async)StopIteration exception

allow yielding exception instances from async generators (previously interpreted them as actual exceptions)

remove accidentally duplicated finaliser code from the coroutine deallocator function as it must get executed earlier than that

add missing preprocessor guard around coroutine dependant code in generic generator utility function

add "--failfast" (stop on first failure) mode to test runner (if supported by Python's unittest module)

refactor "yield from" and "await" nodes a little to provide a cleaner split between them (await nodes are not yield-from nodes)

allow "args" tuple argument to be NULL in __Pyx__Coroutine_Throw() and only create a tuple if necessary

clean copy of the current AsyncGen implementation proposed for CPython 3.6 to make local changes visible

Remove a macro from helper code that I find difficult to read. The result is not perfect, but at least it's more visible what get's assigned to what.

use pre-instantiated strings in __Pyx_setup_reduce() to reduce overhead - we will most likely need and indefinitely store them in the types anyway

This fixes #1798.