Fix missing bound-first-name join in suffix-comma branch

[derek73]

Summary

• parse_full_name() calls _join_bound_first_name() in the no-comma branch (reserve_last=True) and the lastname-comma branch (reserve_last=False), but was missing the call in the suffix-comma branch (the parts[0] path taken when are_suffixes_after_comma is true).
• Result: HumanName("Abdul Salam Hassan, MD") split the bound first name into first='Abdul', middle='Salam', while the equivalent no-comma ("Abdul Salam Hassan MD") and lastname-comma ("Hassan, Abdul Salam, MD") forms correctly produced first='Abdul Salam'.
• Added the missing _join_bound_first_name(pieces, reserve_last=True) call at the suffix-comma call site, matching the no-comma branch's treatment of the same "title first middle last" token sequence.

Why not centralize the join inside parse_pieces()?

Considered moving the join into parse_pieces() itself, keyed off additional_parts_count, so callers wouldn't each need to remember to call it. Rejected: parse_pieces() is called with the same additional_parts_count value both for the post-comma given-names portion (which should get the join) and for the lastname portion in the lastname-comma branch (which must never get it, since it's surname text, not first-name text). Since the same parameter value means different things depending on which slice of the name is being parsed, doing this centrally would need extra context to disambiguate — no simpler than the current explicit per-call-site design, and riskier to get right. Documented this reasoning in AGENTS.md.

Changes

• nameparser/parser.py: added the missing call in the suffix-comma branch.
• tests/test_bound_first_names.py: added regression tests for the suffix-comma form (basic join, join-with-middle, two-token no-join guard).
• AGENTS.md: updated the parse-flow notes to describe all three call sites and why the join isn't centralized.

Test plan

• python -m pytest tests/test_bound_first_names.py -v — all pass, including new suffix-comma tests
• python -m pytest -q — full suite passes (1238 passed, 4 skipped, 22 xfailed), no regressions
• Manually verified all three equivalent input forms now agree: "Abdul Salam Hassan, MD", "Abdul Salam Hassan MD", "Hassan, Abdul Salam, MD" → first='Abdul Salam'

🤖 Generated with Claude Code