The original issue is reported here.
https://discuss.python.org/t/non-optimal-bz2-reading-speed/6869

1. Only BZ2File uses RLock()

lzma and gzip don't use RLock(). It adds significant performance overhead.
When I removed `with self._lock:`, decompression speed improved from about 148k line/sec to 200k line/sec.


2. The default __iter__ calls `readline()` for each iteration.

BZ2File.readline() is implemented in C so it is slightly slow than C implementation.

If I add this `__iter__()` to BZ2File, decompression speed improved from about 148k lines/sec (or 200k lines/sec) to 500k lines/sec.

    def __iter__(self):
        self._check_can_read()
        return iter(self._buffer)

If this __iter__ method is safe, it can be added to gzip and lzma too.