Add a eukaryotic gene caller (TransDecoder/MetaEuk) for transcriptomes/genomes lacking annotation

Description of feature

magmap's only annotators today (Prokka, Bakta) are prokaryote-specific gene callers. There's no equivalent for eukaryotic references -- needed for:

• User-provided eukaryotic genomes/transcriptomes lacking a GFF.
• MMETSP entries, if their own shipped annotation turns out not to be directly usable (see #239, which this would unblock item 3 of).
• Eukaryotic genomes sourced the same way magmap already sources prokaryotic ones -- via a Sourmash index -- which won't come with any annotation at all, same as a locally-provided genome today gets run through Prokka/Bakta.

That third point means this isn't purely a transcript-level (MMETSP) concern with genome support as a hypothetical extension -- a eukaryotic-genome Sourmash-index pathway would need gene calling from day one, the same way the existing prokaryote pathway needs Prokka/Bakta today. Both TransDecoder (transcript-level) and MetaEuk (genome-level) are real, near-term needs depending which reference type magmap ends up supporting first, not a "nice to have later" split.

What's already available (checked nf-core/modules directly, not assumed)

Both tools already have vendored nf-core modules, ready to nf-core modules install:

• transdecoder/longorf + transdecoder/predict -- the standard two-step TransDecoder workflow (find candidate ORFs, then score/select). TRANSDECODER_PREDICT emits .transdecoder.gff3 natively -- when run directly on assembled transcript sequences (not genome-aligned), the GFF3 coordinates are relative to the transcript itself, which is exactly the reference magmap would be using. This looks like it needs little to no extra conversion glue for the transcript-level case (MMETSP, user-provided transcriptomes).
• metaeuk/easypredict -- ab initio eukaryotic gene calling directly on genomic/metagenomic contigs, handling intron/exon splicing. This is the tool for eukaryotic genome references (not pre-assembled transcripts) -- whether user-provided or pulled via a future eukaryotic Sourmash index -- a capability TransDecoder doesn't have, since it only finds ORFs within already-spliced sequences. Its output is .gff -- not confirmed to be strict GFF3, worth checking against magmap's GFF parsers (PROKKAGFF2TSV) before assuming compatibility, unlike TransDecoder's more directly-usable output.

Suggested split

• TransDecoder for the transcript-level case (MMETSP, user transcriptomes) -- pairs directly with #239, likely the first one implemented since MMETSP is the nearer-term concrete goal.
• MetaEuk for the genome-level case (user-provided eukaryotic genomes, or a future eukaryotic Sourmash-index pathway) -- implement whenever genome-level (not just transcript-level) eukaryotic references become the active goal, which may come later than the MMETSP work but isn't lower-priority in principle, just sequenced after it.

Open question

PROKKAGFF2TSV's GFF parsing is reasonably generic (handles standard key=value column-9 attributes, tolerates a couple of naming variants like db_xref/dbxref) rather than tightly Prokka-specific, which is encouraging -- but the exact attribute names TransDecoder/MetaEuk use haven't been checked against what it expects. Worth a real test before assuming either tool's GFF3 output "just works" downstream.

Start at the existing Prokka/Bakta workflow and the PROKKAGFF2TSV parser to understand how annotated references enter downstream processing. Test TransDecoder and MetaEuk outputs against that parser, then define the transcriptome and genome paths and consider the work complete when the selected tool output is accepted downstream with coverage for its GFF attributes.