Fix taxonomy check for kingdom in training script

[edgeemer]

First time doing pull requests so don't get angry at me if something is wrong with the submission :)

Minor fix to sometimes absent “kingdom” flag & TypeError related to non-searchable NCBI species names.

non-searchable NCBI species names: for custom names runs were resulting in TypeError:

funannotate2/train.py", line 353, in train
    if taxonomy["kingdom"] == "Fungi":
TypeError: 'bool' object is not subscriptable

absent “kingdom” flag: for some species (like Paramoeba/Perkinsela) kingdom key can be absent, resulting in KeyError: 'kingdom':
{
"superkingdom": "Eukaryota",
"phylum": "Discosea",
...
}
{
"superkingdom": "Eukaryota",
"phylum": "Euglenozoa",
...
}

[nextgenusfs]

Thanks so much for the fix! This perfectly resolves the TypeError and KeyError issues during training for organisms like Paramoeba where JGI taxonomy lacks a kingdom.

We've merged this PR and will also be applying the same .get() safeguard over in predict.py to ensure the entire pipeline handles these edge cases gracefully. Great catch!