Korean names don't transliterate if they contain punctuation

[scossu]

Reported from Jenna:

In korean_names, S2R:

도, 상록.
to, Sang Nok🜁75ŏk (The result should be: To, Sang-nok. )

도, 상록
to, Sang Rok (The result should be: To, Sang-nok )

도 상록.
To Sangnok🜁-75ŏk (The result should be: To Sang-nok. )

도 상록
To Sang-nok (This is correct.)

It looks like punctuation is interfering with the scanning of the characters.

[kirkhess]

korean_names handles the MARC inverted-name comma inconsistently, and one
case emits an internal assembly marker into the output:

Input	Output
도, 상록.	"" + "is not a recognized Korean name"
도, 상록	"" + same warning
도,상록	To 🜁-75ŏsangnok
도 상록	To Sang-nok ✅

The third case is the bug: 🜁 is INI from
scriptshifter/hooks/korean/romanizer.py:42 ("Initial prefix"), so jamo
assembly markers are surviving into the returned string when the input has a
comma with no following space.

The first two are arguably working as designed — the table matches whole
headings against a name list — but it does mean no MARC-form personal name
matches, since 4XX variants are always inverted (김, 대중, 이, 순신 and
클린턴, 힐러리, 1947- all behave the same). FYI I work around it client-side by
retrying with the punctuation stripped.