bhojpuri (bho): POS model support

I was working on my Bhojpuri project and looked for Stanza support and found there isn't any. bho_bhtb is registered in short_name_to_treebank.py, but bho appears nowhere in default_packages.py — no treebank, pretrain, charlm, or transformer.

UD_Bhojpuri-BHTB ships only bho_bhtb-ud-test.conllu: 357 sentences, 6,665 tokens, no train/dev. It is the same shape mixed_odia_dataset.py handles. I tried it, and it runs on Bhojpuri unmodified, exit 0, both modes:

python stanza/utils/datasets/indic/mixed_odia_dataset.py --mode pos \
  --odia_file $UDBASE/UD_Bhojpuri-BHTB/bho_bhtb-ud-test.conllu \
  --dataset_name bho_bhtb --use_hindi --use_urdu

Odia split: 278 train / 45 dev / 34 test
hindi: 13306 → 1000   urdu: 4043 → 1000

Everything Odia-specific looks cosmetic: the --odia_file arg name, the "Odia split:" print, the or_odtb default. The docstring raises repurposing it for more datasets as future work, but the script already does it — what's left is a rename, not a rewrite.

Happy to do these if useful:

1. Rename to target-neutral args (--target_file, --target_shortname), donors configurable, Odia defaults unchanged — same file, not a per-language copy.
2. Bhojpuri charlm: default_charlms has sd but not bho, and community_oscar_dedup.py documents the Sindhi recipe. I'd check for a bho Community-OSCAR slice and run it.

Questions first:

• Would you take the rename, or prefer separate scripts per language?
• 34 test sentences is too few for a trustworthy number. What evaluation would make Bhojpuri models shippable, or is BHTB too small to bother?
• Odia and Sindhi lean on MuRIL, which I don't think covers Bhojpuri. Does the recipe need the target in the transformer's pretraining data?