Seismic-rna

Bug fixes
- Fixed a bug in `seismicrna.core.batch.index`: calling `np.full(target.max(initial=-1) + 1, -1)` where `target` is a NumPy NDArray of unsigned integer type would implicitly convert -1 to an unsigned integer with the maximum value of its data type (e.g. 4,294,967,295 for a 32-bit integer) and attempt to allocate memory for an enormous array of this size. On some systems, this would cause a crash, and on others would simply waste time allocating the memory.

Performance upgrades
- Mutation data is now processed and saved in a sparse format that tracks only the mutated positions. Since mutations make up only 1 - 5% of most datasets, the sparse format is more storage-, memory-, and time-efficient.
- Batches are now saved in Brotli-compressed pickle ("Brickle") files, which requires less storage and allows more types of data to be saved than the previous Parquet and gzip-compressed CSV formats.

New features
- The `table` step has been sped up via the sparse data format, and computing all fields is nearly as fast as computing one field. Thus `table` now computes all fields automatically (the option to compute select fields has been removed).

Bug fixes
- When running `align` on demultiplexed FASTQ files, one report file is now generated for each FASTQ file, rather than all FASTQ files for each sample writing to and overwriting one report file.
- When running `relate` on multiple samples that are aligned to the same set of references, every BAM/CRAM file from every sample is processed instead of only one sample BAM/CRAM file for each reference.
- When running `fold`, misformatting of the RNAstructure `Fold` command has been fixed.

Internals
- The `core` modules have been refactored into a group of subpackages, each with their own modules.
- The `all` subcommand has been moved from the `main.py` module to its own subpackage.
- The mutation calling and counting routines in the modules `seismicrna.core.bitcall` and `seismicrna.core.bitvect`, respectively, have been rewritten and replaced with `seismicrna.core.rel.pattern` and `seismicrna.core.batch.accum`.
- The unique read finding algorithm has likewise been rewritten and moved to `seismicrna.cluster.uniq`.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.8.0...v0.9.0

- Align now generates CRAM files with minimal headers instead of BAM files with full headers so that large FASTA files and large FASTQ files eventually require less storage space.
- Align has been re-implemented as two shell pipelines instead of as a series of separate commands glued together with Python, to make it run faster and require less storage of temporary files.
- A new function for parsing only the names of references in FASTA files (if the sequences are not needed) is based on grep and runs several times faster on large files than does the Python-based function for parsing both names and sequences.
- The ambiguous nucleotide "N" is now supported in both reference and read sequences (previously, neither).
- Unit tests have been updated to handle N in DNA and RNA sequences.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.7.1...v0.8.0

- Added Hatch targets to `pyproject.toml`
- Updated documentation

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.7.0...v0.7.1

- Output directories are now organized as `out/sample/step/ref` instead of `out/step/sample/ref`.
- Documentation has been partially updated.
- More unit tests have been added.
- A release schedule has been added.
- The --min-mapq option has been added.
- Log messages are color coded.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.6.2...v0.7.0

- Fixed bug with min_nmut_read not being accepted by `main.run()`

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.6.0...v0.6.2