Fixes #840

What's the issue?

When snp_allele_frequencies (or snp_effects) is called on a transcript that has only a single child — for instance a single-exon non-coding transcript — Annotator.get_children() in veff.py returns a pandas Series rather than a DataFrame. The downstream .sort_values(start) call then blows up with a TypeError, because Series doesn't have the same sort_values semantics when columns are referenced by name in that context.

As @jonbrenas noted in the issue, the deeper question is what to do with transcripts that lack coding regions entirely. The effects we compute (synonymous, start/stop lost, etc.) don't make sense for non-coding transcripts, so a clear error is more helpful than a cryptic crash.

What I changed

malariagen_data/veff.py

• get_children() — when DataFrame.loc[] matches a single row, pandas returns a Series. I added a check: if the result is a Series, convert it back to a single-row DataFrame with .to_frame().T and preserve the index name. This way the rest of the pipeline (sorting, column filtering, .itertuples()) works as expected regardless of how many children the transcript has.

• get_effects() — after extracting CDS, UTR, and exon children, I added a guard that checks whether the transcript has any coding features (CDS or UTR). If not, it raises a ValueError with a message explaining that variant effect annotation only applies to protein-coding transcripts and that the annotation may be incomplete. This replaces the previous behavior of silently producing incorrect results or crashing further downstream.

How I tested

Wrote a standalone test script that exercises three scenarios:

1. Single-child transcript — confirms get_children() returns a DataFrame and sort_values(start) works
2. Multi-child transcript — confirms the fix doesn't break the normal case
3. Non-coding transcript — confirms the new ValueError is raised with an informative message

All three pass.