Timing-sensitive threaded tests starve each other on loaded runners #28

Found in #27: adding a few extra threaded tests made unrelated sleep-based tests (test_stream_buffer, test_reader_works_without_context_manager, test_next_returns_none_when_no_data_but_running) fail consistently in CI, and master fails a different one of them when pinned to one core (taskset -c 0). They sleep a fixed time then expect data. They should poll for the condition with a deadline.