Fix BGZF truncation at embedded EOF markers

[KimYannn]

No description provided.

[aalnawfal1992]

Tested this patch against #711 on real data It fixes the problem and reproduces pre regression behaviour exactly. Built master and this PR from source (Ubuntu 24.04, isa-l / libdeflate / libhwy from apt).

1. Real data

A WES sample straight off bcl convert, not manually concatenated The R1 file contains two BGZF streams and 24,910,823 read pairs (verified with zcat | wc -l).

build	read pairs seen
master	24,854,834	stops at the end of stream 1
this PR	24,910,823	reads the whole file

The master figure is exactly the number of read pairs in the first BGZF stream, confirming it halts at the first EOF block.

2. Minimal reproducer

Two samll files of 200 read pairs each, compressed separately and concatenated, so the correct answer is 400:

bgzip -c a_R1.fastq > a.bgz          # 200 pairs
bgzip -c b_R1.fastq > b.bgz          # 200 pairs
cat a.bgz b.bgz > merged_R1.fastq.gz # 400 pairs

build	read pairs seen	expected
master	200	400
this PR	400	400

A single stream bgzip file (200 pairs, no embedded EOF marker) reads correctly as 200 on both builds, so the patch does not regress the normal case and does not hang. That was worth checking, since break -> continue means the trailing EOF marker no longer terminates the loop and markDone() is not called on that path.

3. Equivalet with v1.1.0

v1.1.0 is the last release before the regression. Same file, same flags. These counts are R1+R2 combined, as reported in the JSON:

metric	v1.1.0	this PR
reads before filtering	49,821,646	49,821,646
reads after filtering	49,446,918	49,446,918
low_quality	366,304	366,304
too_many_N	4,422	4,422
too_short	3,996	3,996
adapter trimmed reads	125,966	125,966
adapter trimmed bases	6,376,503	6,376,503

Identical on every metric, and the final read in the output file is the same in both.

Read order and pairing are also preserved: the first read of stream 2 appears in the output immediately after the last read of stream 1, and R1/R2 name order is identical.

Note; Concatenated plain gzip was never affected, only BGZF

[omegahh]

Looking forward to having this PR merged.

[sfchen]

thanks, merged.