AdapterRemoval v3.0.0-alpha2

This is the second alpha release of AdapterRemoval v3. It is the intention that
a third alpha release, or the final 3.0 release, will follow within the next
couple of months.

As with alpha 1, changes that affect how AdapterRemoval is used (e.g. by
removing options) or that result in different output compared to AdapterRemoval
v2 are marked with the label "[BREAKING]".

In addition to changes listed below, this release includes increased throughput
thanks to improved parallelization of various steps in internal pipeline,
support for AVX512 and general improvements to the SIMD alignment algorithms,
loop unrolling of non-SIMD alignments to significantly increase throughput when
SIMD is not available, and a significant decrease in the number of allocations
to decrease overhead.

This release requires a compiler with support for c++17 and libdeflate is now a
mandatory dependency.

Draft documentation is available here and a pre-compiled binary for x86-64
Linux systems is attached below.

Added

• Added support for converting (U)racils in input data to T(hymine) via the
  --convert-uracils flag.
• Added support for replacing IUPAC-encoded degenerate bases with Ns via the
  --mask-degenerate-bases flag.
• Added support for writing output in SAM/BAM formats, with optional
  user-supplied read-group information.
• Added support for alignments using AVX512 instructions. AVX512 support only
  available when AdapterRemoval is compiled with GCC v11+ or Clang v8+.
• Added support selecting output file formats via the file extension and via
  the --out-format option. A corresponding option, --stdout-format was
  added to select the format for data written to STDOUT.
• Added support for reading from STDIN or writing to STDOUT when '-' is used as
  the filename, as an alternative to using /dev/stdin or /dev/stdout.
• Added dedicated threads solely for writing output data. This allows compute
  threads to work at full capacity, as long as the destination can consume
  written data fast enough. This may result in CPU utilization exceeding
  --threads by a couple of percent.
• Added support for setting DESTDIR when running make install.
• Added --licenses flag for displaying licenses of 3rd party code used by /
  incorporated into AdapterRemoval.
• Added --simd option allowing the user to select the specific SIMD
  instruction set they wish to use.
• Added Containerfile for building static binaries using alpine/musl.

Changed

• [BREAKING] Changed the default --mm/--mismatch-rate from 1/3 to 1/6,
  in order to decrease the false positive rate, in particular for read merging.
• [BREAKING] Default to writing gzip-compressed FASTQ files; output written
  to STDOUT is uncompressed by default.
• [BREAKING] Discarded reads are no longer saved by default.
• [BREAKING] Output files for discarded reads and singleton (orphan)
  paired-end reads are only created if filtering is enabled.
• [BREAKING] The --basename / --out-prefix no longer defaults to
  your_output. Instead the user is required to set at least one --out-*
  option.
• [BREAKING] Merged --identify-adapters and --report-only commands. The
  adapter sequence is presently only reported in the HTML report, but will be
  added to the JSON report following some planned changes.
• [BREAKING] Reverted --min-complexity being enabled by default.
• Increased the default --threads value to 2.
• A number of command-line options were renamed for consistency; use of the old
  names is still supported, but will trigger a warning message.
• Re-organized compression: level 1 is streamed using isa-l, while levels 2-13
  correspond to libdeflate levels 1 to 12.
• Changed the default compression level to 5 on the new scale (libdeflate level
  4); this results in a ~40% increase in throughput at the cost of roughly ~3%
  larger output files.
• Setting an --out-* option in demultiplexing mode overrides the basename /
  prefix for that specific output type.
• Add smoothing to GC values calculated for the GC content curve, to account
  for the fact that possible GC% values are unevenly distributed depending on
  the read length.

Removed

The following changes are all [BREAKING] as described above:

• Removed support for original merging algorithm has been removed. The
  --merge-strategy additive method produces very similar, but slightly more
  conservative scores.
• Removed the ability to randomly sample a base if no best base could be
  selected in case of mismatches. Such bases are now changed to N, while both
  methods assign a Phred score of 0 (!).