Fix paired-end output, rename fragments to segments, expand the digest report

Fixes #3 #2

Fixed

--gzip silently dropped records over ~8KB, leaving R1 and R2 different lengths. Only
0.23% of pairs were correctly mated; hifiasm was reading mismatched pairs.
The default uncompressed writer had the same unchecked-write bug.
R1 and R2 carried /1 and /2, so the two files disagreed on every pair name.
-l was not honored for R2, which came out shorter than asked for. Low values of -l
produced 1bp reads.
R2 was never reverse-complemented on the default path.
Report sections were missing under --fast.
Yield figures ignored reads that failed the filters, so the bases did not add up.
reads_in counted BAM records the statistics never saw.
Out-of-range cut_offset in process_reads_custom caused an underflow and overlapping segments.

-l default is now 60, was 20. hifiasm drops a pair outright if either mate is under
31bp, so shorter mates were written and then discarded.
--revcomp-r2 is now off by default. R2 keeps the orientation it was sequenced in.

--min-fragments and --min-frag-len are now --min-segments and --min-segment-len.
Short -m and -l are unchanged.
Python keyword arguments and _stats.json keys renamed to match.
--revcomp-r2 no longer implies --no-strip-overhang; the two options are independent.