Fix HPO parsing for OBO 1.4 - round 2

[EddieLF]

Follow up to #288

The problem

The regex assumed {modifier} comes after ! comment, but OBO 1.4 spec puts the modifier before the comment, and the real HPO file uses the standard ordering. With a line like is_a: HP:0032162 {xref="PMID:31677808"} ! Phenotypic abnormality, the anchored-to-end regex didn't match, split(' ! ')[0] returned HP:0032162 {xref="..."}, and that string was stored as the parent_id, leading to errors.

The fix

• re.search(r'HP:\d{7}', value) extracts the parent id directly, regardless of where (or whether) modifiers/comments appear around it.
• Raises with the offending line if no HPO id is found, so a malformed is_a: line surfaces a clear error instead of silently storing garbage as parent_id (which is what bit us in the last PR).
• update_hpo_tests.py — added a new fixture term HP:9999003 using the OBO 1.4 standard ordering ({modifier} ! comment) so this case is covered going forward; bumped record counts to 8.

[MattWellie]

Oof, so it was just bad luck that the original failure we spotted was ordered a certain way?