HumanName.initials() re-derives the conjunction decision instead of honoring the parse's tags: john e smith gives j. s. from the facade and j. e. s. from parse() #528

Since PR #527, the core and the v1 facade split on which single letters are connectives. parse("john e smith").initials() gives j. e. s. (the e is read as an initial in a name written wholly in one case, rules.md#P3), while HumanName("john e smith").initials() still gives j. s.. The split runs both ways: parse("JUAN Y GARCIA").initials() gives J. G. and the facade gives J. Y. G.; parse("JUAN GARCIA Y LOPEZ").initials() gives J. G. L. and the facade J. G. Y. L..

Mechanism. _facade._process_initial decides "conjunction" from vocabulary plus initial shape on the part's raw text (self._is_conjunction(part) and not _render._INITIAL.fullmatch(part)), never from the parse's conjunction / initial tags. #462 fixed one half of that (a bare capital or dotted letter is an initial) but the facade still cannot see the one-case fork, and capitalize() already delegates to the parsed view while initials() does not.

Fix shape, the same as R4's case repair. Where a part maps to a parsed token, read its tags: a conjunction-tagged word contributes no initial, an initial-tagged one contributes; fall back to vocabulary and shape only for spliced raw text (hn.middle = "...") that no token backs. That restores the #458 principle for the last derived view that re-decides from the word.

Pinned today as a deliberate divergence by tests/v2/test_render.py::test_facade_initials_do_not_yet_follow_the_one_case_fork; recorded at decisions.md#P3 as an Accepted consequence of #527. Closing this moves the facade's _initials in the differential at every baseline, so it needs its own ledger rules (the core's moved in #527 at 2.0.0–2.3.0 only).