`abdul Jr Smith Berg` reads given `abdul Jr` — P5's join takes a suffix as "the word after" the bound word

[derek73]

A bound given-name word joins whatever piece follows it, suffix included:

abdul Salam Smith Berg     ->  given='abdul Salam'  middle='Smith'  family='Berg'
abdul Jr Smith Berg        ->  given='abdul Jr'     middle='Smith'  family='Berg'   suffix=''
abdul PhD Smith Berg       ->  given='abdul PhD'    middle='Smith'  family='Berg'   suffix=''
Berg, abdul Jr Smith       ->  given='abdul Jr'     middle='Smith'  family='Berg'   suffix=''

and where the suffix is the split credential, the bound word joins into it and becomes part of the suffix:

abdul Ph. D. Smith Berg    ->  given='Smith'  family='Berg'  suffix='abdul Ph. D.'

Why

rules.md#P5 says the join "joins the word after it into one given name" and "needs a name word to spare". The reserve honours the second half — it counts only non-suffix pieces, so abdul Jr Jones declines — but the join itself never asks whether the piece it is about to take is a name word. In _group.py the bound-given block checks absorbs_marker (#411: "a marker is not a name word") and nothing else, then merge(first_name_k, first_name_k + 2).

The credential shape is worse because merge() unions piece tags: the "Ph. D." piece carries suffix from the ph-d merge, so the joined abdul Ph. D. piece is a suffix piece and assign routes the bound word to the suffix field.

Scope

• The Jr/PhD shape is pre-existing: 1.4.0 reads abdul Jr Smith Berg as first abdul Jr too. Parity, not a regression.
• The split-credential shape is a 2.0 regression: 1.4.0 read abdul Ph. D. Smith Berg as first abdul Smith, suffix Ph. D. (v1's fix_phd extracted the credential before parsing, so position never mattered). 2.x reads suffix abdul Ph. D..
• No differential corpus name has a bound word directly followed by a suffix (36 carry a bound word; none with this shape), so the gate cannot see either half. Surfaced by fix(group): run the maiden marker pass before every join (#412, #417, #418) #420's re-pin of abd née Jones Jr Smith Berg, which reads given abd Jr because abd Jr Smith Berg does; tests/v2/cases.py::bound_given_join_sees_only_the_surviving_name pins today's reading and would move.

Fix

Decline the join when the piece it would absorb is a suffix piece, the same way it already declines for a marker — one test beside absorbs_marker. That makes the rule read as it is stated: the join takes a name word, and the reserve and the join then agree about what one is. abdul Jr Smith Berg would read given abdul, middle Jr Smith, family Berg (the inner-suffix-to-middle placement abdul Jr Jones already has), and abdul Ph. D. Smith Berg would read given abdul, middle Smith, family Berg, suffix Ph. D. — 1.4.0's reading restored.

P5's statement would gain a clause: "The join never absorbs a marker (M2) or a suffix (S1) — neither is a name word." Worth a row each in the case table and an example line under P5, since the differential corpora cannot carry the shape.