test suite: find a way to prevent unclean reactor errors causing subsequent test failures

[davidsarah]

For example, in the 666-accounting branch, if test_mutable.Filenode.test_modify_backoffer is reenabled, it will fail with an unclean reactor error and subsequent tests will time out.

Probably this is due to some non-obvious sharing of global state between tests (it doesn't always happen that an unclean reactor error causes subsequent timeouts).

In any case, it makes it difficult to run the test suite the whole way through when some tests cause this problem. One rather brute-force solution would be to exit on an unclean reactor error and restart the test process from the next test. I can't see a way to do that without forking trial, though. Another possibility is to try to use ​twisted.test.proto_helpers.MemoryReactor.