We have no consolidated, outdir-scoped protein FASTA output

Description of feature

Prokka and Bakta already produce a per-genome amino-acid FASTA (`*.faa.gz`), but it only ever lands scattered inside `--prokka_store_dir`/`--bakta_store_dir` -- a persistent cache directory reused (and potentially shared) across pipeline runs, not scoped to `outdir`. There's currently no single "here are all the proteins from this run's genome set" file anywhere under `outdir`.

This came up thinking about chaining magmap into downstream protein-annotation pipelines (e.g. nf-core/proteinannotator, which just wants a per-sample `id,fasta` samplesheet of amino-acid FASTA) -- whether by hand, by a metapipeline tool like nf-meta, or anything else. Right now that would mean manually globbing/concatenating `.faa.gz` files out of a store directory that isn't even guaranteed to correspond 1:1 with a single run's genome set.

A consolidated protein FASTA under `outdir` (e.g. alongside the existing `summary_tables/` outputs, or wherever makes sense) would fix this on its own merits, independent of any specific downstream tool -- similar in spirit to how `CAT_FNA`/`CAT_GFF` already concatenate the nucleotide/GFF side per genome set.

Complication to design around

Not every genome in a genome set goes through Prokka/Bakta: genomes supplied with their own GFF (via `--genomeinfo`) skip annotation entirely and so never get a Prokka/Bakta-produced `.faa`. A naive "concatenate all `.faa.gz` files" step would silently omit those genomes' proteins from the consolidated output. Worth deciding up front whether that's acceptable (documented limitation) or whether it needs translating CDS features from the user-supplied GFF + genome FASTA to fill the gap (bigger scope).

Not tied to any milestone yet -- flagging for later triage.

Start by tracing the existing CAT_FNA/CAT_GFF outputs and the Prokka/Bakta annotation paths under outdir and their store directories. Define the consolidated amino-acid FASTA location and verify how genomes supplied with --genomeinfo are handled; done means one run-scoped output covering the agreed genome set, with tests or documentation for any deliberate omissions.