packed-seq 4.4.2

# Changelog

<!-- next-header -->

## git

## 4.4.2
- Bugfix on 32-bit platforms where reverse-complementing k-mers used `usize::BITS` instead of `u64::BITS`.

## 4.4.1
- Improve docs

## 4.4.0
- Add `PackedSeq{Vec}Base::from_raw_parts` to construct directly from a `Vec` or slice.
- Add `PackedNSeqVec::from_ascii_and_quality` that masks out low-quality bases.
- Add `PackedNSeqVec::from_{fastx,fastq_with_quality}` behind `file_io` feature
  (for `needletail` dependency).
- Add experimental `PackedEfNSeq` that stores `N` positions in an EF vector,
  behind `ef_seq` feature flag (for `sux` dependency).

## 4.3.0
- Add `unpack_kmer_to_vec(kmer: u64, k) -> Vec` and variants.
- Add `pack_kmer_lossy(kmer: &[u8]) -> u64` and variants.
- switch to `ensure_simd`

## 4.2.0
- Perf: improve NEON codegen by avoiding `u32x8::splat`.
- Perf: reuse thread-local allocations between invocations
- Feat: `par_iter_bp_*_with_buf` versions that allow passing in buffers from the
  stack, which can be slightly faster than thread-locals.
- Perf: all `PackedSeq` slices have padding to allow faster out-of-bounds reading.
- Perf: Rewrite `par_iter_kmer_ambiguity` to use a delayed iterator.
- Feat: Add `PackedNSeq::slice(Range<usize>)` utility.

## 4.1.1
- Feat: `PackedNSeqVec::{default,clear,push_ascii}`.
- Feat: `BitSeqVec::par_iter_kmer_ambiguity` supports `k` up to `96`.

## 4.1.0
- Perf: Further optimizations to `BitSeq::from_ascii` and `PackedSeq::from_ascii` for
  both AVX and NEON. (Fewer out-of-bounds checks, NEON optimizations to work
  around lack of `movemask`.)
- Perf: Add small benchmarks (in `src/test.rs`) for `{Bit,Packed}Seq::push_ascii` and `PackedSeq::par_iter_bp`.
- Perf: More `#[inline(always)]` for better performance with `lto="thin"`.
- Feature: Add `BitSeqVec::random`.
- Feature: Add `PackedNSeq` as a wrapper around `(PackedSeq, BitSeq)` to hold
  2+1-bit encoded data, with the first 2 bits encoding ACGT, and the last bit
  indicating non-ACGT IUPAC characters (such as N, R, Y, ...).
- Feature: Add `PackedSeq::par_iter_bp_delayed_with_factor` (mirring
  `par_iter_bp_delayed_2_with_factor`) for use in `seq-hash`.
- Bugfix for repeated `BitSeq::push_ascii`.
- Bugfix for `BitSeq::to_revcomp` (which was still 'complementing' instead of only reversing).

## 4.0.1
- Feature: optimize `BitSeq::from_ascii` for NEON.
- Bugfix: 1-bit `reverse` was also doing 'complement'.

## 4.0.0
- **Feature**: Add `BitSeq` and `BitSeqVec` as 1-bit encoded 'sequences' that by default
  indicate the positions of ambiguous (non-ACTG) characters.
- **Feature**: Add `BitSeq::par_iter_kmer_ambiguity` that returns for each kmer
  whether it contains an ambiguous base.
- **Breaking**: Encapsulate parallel iterators in new `PaddedIt { it, padding: usize }` type with `.map`, `.advance`, `zip`, and `.collect_into` functions.
- **Breaking**: Make `delay` passed into `par_iter_bp_delayed` a strong type `Delay(pub usize)` to
  reduce potential for bugs.
- Make `intrinsics::transpose` public for use in `collect_and_dedup` in `simd_minimizers`.

## 3.2.1
- Add `Seq::read_{revcomp}_kmer_u128` with more tests
- Fix bug in `revcomp_u128`

## 3.2.0: yanked
- Add `Seq::as_u128` and `Seq::revcomp_as_u128`
- `revcomp_u128` had a bug; fixed in 3.2.1

## 3.1
- `PackedSeqVec::{from,push}_ascii` now silently convert non-ACGT characters to
  values in `0..4`, instead of inconsistently only panicking for non-ACGT near
  the sequence start and end.

## 3.0
- Speed up `PackedSeqVec::push_ascii()` on ARM
- Add `Seq::to_revcomp()`.
- Add `PackedSeqVec::clear()`.
- `PackedSeq::as_u64` now supports k up to 32, instead of only up to 29. (Fixes #2.)
- `PackedSeqVec::get` now uses checked indexing, to prevent unsafe out-of-bounds access.
- **Breaking**: `PackedSeqVec` and `PackedSeq` fields are now private, to uphold internal
  padding guarantees.
- Add convenience methods `{Seq, SeqVec}::{read_kmer, read_revcomp_kmer}` as
  shorthands for `.slice(pos..pos+k).as_u64()`.
- Deprecate `to_word` and `to_word_revcomp`. Instead `as_u64` and
  `revcomp_as_u64` are added. ('word' is not very clear, and `u64` makes more
  sense for kmers than `usize`.)
- Support `par_iter_bp` for sequences that do not start at byte offsets.
- Add `packed_seq::revcomp_u64()` and `Seq::revcomp_as_u64()` for reverse
  complementing kmers.
- Make `random` dependency optional but enabled by default (so it can be
  disabled for wasm backends).

## 2.0
- Change from tuples of (simd iterator, scalar tail iterator) to returning only a
  simd iterator and `padding: usize` that indicates the number of padded elements
  at the back that should be ignored.

## 1.0
- Initial release.