Optimizes iter_lines handling long lines

[mfulgo]

In the previous implementation of iter_lines, lines that spanned chunks could cause a quadratic iteration of the line data. This could result in a much slower read time for files with long lines or lines significantly longer than the chunk size used when reading them.

To avoid the issue, the StreamingBody now collects the line data and only constructs the line (copying the bytes into a new buffer) when it encounters a line ending.

Closes #2774

[mfulgo]

I wasn't quite sure how you all might want a performance test written for this, so I just added and ran the following unit test before and after the change:

    def test_streaming_line_iterator_long_lines(self):
        line = b'1234567890'*1000000
        body = itertools.repeat(line, 1)
        stream = response.StreamingBody(BytesIO(b'\n'.join(body)), content_length=(1*10*1000000)+1-1)
        start = perf_counter()
        for l in stream.iter_lines():
            self.assertEqual(l, line)
        raise RuntimeError(f"It took {perf_counter()-start:.3f} seconds")

Prior to this commit, it took around 23 seconds to run. After the change, it consistently ran in about 0.009 seconds.