Too many extra reads are returned

[kemp-google]

The specification allows the server to return more reads than were requested. The current implementation errs too much on that side of things.

As an example, from Jerome's test harness over ENCFF284YOU:
2017-10-13 10:41:27,515 Downloaded 95.6 KiB in 0.906 seconds
2017-10-13 10:41:27,559 Indexed BAM file in 0.004 CPU seconds
2017-10-13 10:41:27,582 Filtered 0 reads
2017-10-13 10:41:27,583 Checked reads in 0.023 CPU seconds
2017-10-13 10:41:27,585 Downloaded 166 reads with 1119 extra
523 queries downloaded 652.8 MiB of data at an average of 0.91 Mib/s

We may not want to get to 0 extra reads for server side efficiency but we should be able to do better than ~6x too many.