Genomad

Changed
- Add installation instructions using Pixi.
- Use raw strings for the regular expression in `utils.natsort`.
- Update the version requirements for `keras` to `>=3` and `tensorflow` to `>=2.16`.

Fixed
- Write the `min_number_genes` value to the parameters JSON file of the `summary` module.
- Set maximum `tensorflow` version to below `2.16`.

Changed
- Set the `break_on_hyphens` parameter of the `textwrap.fill` function to `False` to prevent line breaks at `-` characters. This ensures that sequences with gaps in FASTA files generated using `Sequence.__str__()` maintain consistent line width.
- Compare `Enum` by identity in the `open_file` function.

Added
- Added the `--min-number-genes` parameter to the `summary` module. This parameter allows users to set the minimum number of genes a sequence must encode to be considered for classification as a plasmid or virus. The default value is `1`. When `--conservative` is used, this parameter is set to `1`. When `--relaxed` is used, this parameter is set to `0`. This filter has no effect if the `annotate` module is not executed.

Changed
- Added a hyperlink to the official documentation in the help dialogue.
- The virus taxonomic lineage is presented using a fixed number of fields separated by semicolons (`;`). As a result, for genomes that could not be assigned to the family level (the most specific taxonomic rank), there will be trailing semicolons at the end of the lineage string.
- Do not apply the gene-based post-classification filters when the `annotate` module is not executed.
- Set the default value of `--min-plasmid-marker-enrichment` to `0.1`.

Fixed
- Set maximum `keras` version to below `3.0`. This prevents errors due to incompatibility with `keras >=3.0`, such as the `shape` parameter not accepting an integer as input.

Fixed
- Set the `CUDA_VISIBLE_DEVICES` environment variable to `-1` in `nn_classification`. This fixes a bug where the `nn_classification` module would fail to run when a GPU was available and the input had a single sequence.

Fixed
- Fixed the parsing of MMseqs2 integrase output to extract only the gene accession, rather than the entire header. This addresses a bug introduced in version 1.5.2, where the integrase gene accession was not accurately parsed because the entire header was extracted. As a result, the `find-proviruses` module can now properly add integrases to gene tables and extend boundaries using integrase coordinates.

Changed
- Replace ambiguous variable name in `read_fasta`.
- Define name `current_contig` at the beginning of `_append_aragorn_tsv`.