`initials()` drops the accent from a decomposed (NFD) name: `émile zola` gives `e. z.`

[derek73]

An initial taken from a name typed in decomposed form (NFD, which macOS file names and some databases produce) loses its accent:

>>> import unicodedata
>>> parse(unicodedata.normalize("NFD", "émile zola")).initials()
'e. z.'
>>> parse("émile zola").initials()
'é. z.'

HumanName(...).initials() does the same. The parse itself is right: roles and tags for the NFD spelling equal the NFC ones.

Cause. nameparser/_render.py's initials() takes t.text[0], a token's first code point. In NFD that is the base letter e, and the combining accent (U+0301) that follows it is dropped.

Reach. Unlike capitalized(), initials() is a compared surface: the differential gate checks it under _initials (#484). The rules corpus has held two NFD names since #584, but both start with an unaccented letter (josé garcía), so the gate doesn't trip.

Options.

1. Take the first letter plus the combining marks after it (the _past_marks helper Fix case repair splitting a decomposed (NFD) word at its accent #584 added to _render.py). The output keeps the input's form, matching how case repair now handles decomposed text (rules.md#R4, decisions.md#R4 2026-10-02).
2. Compose the initial to NFC. Simpler, but the output's form would then differ from the input's. R3 states no "nothing but X" rule here, so this is a choice rather than a violation.

Either way, a rules.md#R3 example line in NFD should pin it. Found while fixing #542; recorded as out of scope in decisions.md#R4's #542 bullet.