feat: add --ko_input option to skip DIAMOND search using pre-computed KO annotations

[ynishimuraLv]

Motivation

DIAMOND-based KO annotation may fail for highly divergent genomes with very fast evolutionary rates (i.e., Candidatus Sukunaarchaeum mirabile), where sequence similarity to reference databases is too low to obtain reliable hits.
This option allows users to provide KO annotations generated by other tools and still leverage CheckM2's quality prediction workflow.

🤖 Developed with assistance from Claude Code

Summary

Adds a new --ko_input option to checkm2 predict that allows users to skip the DIAMOND blastp step by providing
pre-computed KO annotations.

Changes

• --ko_input : Path to a folder containing per-genome KO annotation files (CSV format with gene_id and
  ko columns)
• --input: When --ko_input is specified, --input should point to a folder of protein (amino acid) FASTA files
  instead of genome sequences (Prodigal is skipped)
• Default file extension is set to .faa when --ko_input is used

Annotation file format

Each file in the --ko_input folder should be named {genome_name}.{any_extension} and be a CSV with the
following format:

gene_id,ko
protein_001,K00844
protein_002,K12407
protein_003,
protein_004,K00001

Validation

• All gene_id values in the annotation file must be present in the corresponding protein FASTA file; genomes
  that fail this check are skipped with an error log
• Genes absent from the annotation file (unannotated) are permitted — a warning is logged and prediction
  proceeds
• --ko_input and --resume cannot be used together