Modified base parsing failure failure

[cjw85]

The functions bam_mods_at_next_pos and bam_next_basemod fail to detect when and MM tag is corrupt for reverse-mapped reads.

For example, the records:

@SQ     SN:ref  LN:64
reverse 16      ref     1       255     21M     *       0       21      GGGGGGGTCTCTAACGACCAA   *       MM:Z:T+T?,5,0,0,0,0;    ML:B:C,182,3,4,3,3
forward 0       ref     1       255     21M     *       0       21      TTGGTCGTTAGAGACCCCCCC   *       MM:Z:T+T?,5,0,0,0,0;    ML:B:C,182,3,4,3,3

contain malformed MM tags. The queries contain only 5 T bases (A for reverse record) but the MM tag informs that the first modification is located after skipping over the first 5 Ts (As from the right).

Putting these records through the test program test/test_mod in htslib yields:

0	G
1	G
2	G
3	G
4	G
5	G
6	G
7	T
8	C
9	T
10	C
11	T
12	A	T+T3
13	A	T+T3
14	C
15	G
16	A	T+T4
17	C
18	C
19	A	T+T3
20	A	T+T182
---
Present: T
12	A	T+T3
13	A	T+T3
16	A	T+T4
19	A	T+T3
20	A	T+T182

===

0	T
1	T
2	G
3	G
4	T
5	C
6	G
7	T
8	T
9	A
10	G
11	A
12	G
13	A
14	C
15	C
16	C
17	C
18	C
19	C
20	C
---
Present: T
[W::bam_next_basemod] MM tag refers to bases beyond sequence length

It is tripping up only on the forward record, not the reverse record.

Escalated from: epi2me-labs/modbam2bed#21