Sequali

+ Sort modules for paired end reports in the same order as single end reports.
For example, the sequence length distributions for read 1 and read 2 are now
right after each other.
+ Add common human genome repeats and Illumina poly-G dark cycles to the
overrepresented sequences database.
+ Illumina adapter trimming sequences were added to the contaminants database
as these were missing from the UniVec database.
+ Sequence identity, rather than kmers matched is shown as a metric for
similarity in the overrepresented sequences table.
+ Overrepresented sequence classification now uses stable sorting to ensure
the classification results are the same on each rerun.
+ Overrepresented sequences are now classified using Smith-Waterman alignment
and sequence identity.
+ Fix an off by one error in the insert size metrics that was triggered for
insert sizes larger than 300 bp.

------------------
+ Properly name percentiles as such in the sequence length distribution rather
than using N50 nomenclature which is not correct.
+ Fix a bug where BAM files with missing quality sequences were inproperly
handled.
+ Update internal UniVec database to version from November 21st 2023.

------------------
+ Fix a memory leak that occurred in Python 3.12 due to a refcounting API
change.

------------------
+ Make figure IDs reproducible across HTML reports.
+ Fix a bug where the average phred score per read would be rounded, not
floored. This would lead reads with a phred score such as 9.7 to be counted
towards the Q>=10 results.
+ Replace some of the hand vectorized code with more generic code that can be
automatically be optimized by the compiler. This should make things faster on
Windows and ARM64 platforms. This also means results should be consistent
across platforms and no longer depend on the presence of vector instructions.

------------------
+ Make overrepresented sequences table scrollable and smaller so it is easier
to skip over when lots of entries are found.
+ Overrepresented sequence analysis now only counts unique fragments per read.
Fragments that are duplicated inside the read are counted only once. This
prevents long stretches of genomic repeats getting higher reported
frequencies.
+ Speed up sequence identity calculations on AVX2 enabled CPUs by using a
reverse-diagonal parallelized version of Smith-Waterman.

-----------------
+ Fix an issue where the insert size metrics module would crash when no
adapters where present.