tests: Try to run GLib-based tests under prove(1)
This gives us better diagnostics on failure, and matches what we do in the pressure-vessel subdirectory and for scripts.
tests: Run with G_MESSAGES_DEBUG=all if GLib is too old for TAP
This provides better diagnostic information on failure.
Mostly untested (perhaps ironically!) but might help when diagnosing test failures, particularly in a non-interactive autobuilder environment.