Fix bugs in mpileup.

[alumi]

Summary

This PR fixes some problems in mpileup command.

Fixes

• Fix: may cause StackOverflow when piling up long sequence.
• Fix: quality score may be incorrect for bases with complex cigar.
• Fix: output deletions.
• Memoize parsing cigar string into sequence of indices. Parsing itself also get slightly faster.

Known issues

There is a difference between cljam and samtools for adjacent indels.

input SAM

@SQ SN:ref  LN:20
r001 0 ref 2 60  1M5D5I  * 0 0 ATGCAT  IIIIII

samtools 1.3.1 output

ref	2	N	1	^]A-5NNNNN	I
ref	3	N	1	*	I
ref	4	N	1	*	I
ref	5	N	1	*	I
ref	6	N	1	*	I
ref	7	N	1	*+5GCATC$	I

cljam output

ref	2	N	1	A-5NNNNN	I
ref	3	N	1	*	~
ref	4	N	1	*	~
ref	5	N	1	*	~
ref	6	N	1	*	~
ref	7	N	1	*+5TGCAT	~

I'm not certain about the spec, but I guess the last pileup should be *+5TGCAT$.
The last C at 7th base is an artifact from first nibble of quality IIIIII caused by buffer overrun.
Note that we're omitting ^, mapq and $ to mark start / end of reads.

[totakke]

LGTM in this fix domain.

[totakke]

The pointed INDEL problem has existed since before. It should be noted in cljam issue.

[alumi]

Thank you. I opened an issue #28.