[ronaldtse]

Summary

Repairs the last corpus embedded-test failures across five maps (batch B of two; batch A covers ban/mni/bod/lif separately). Every change is source-anchored - the cited page or note from each system's reference document.

Per map (all fixes verified 85/85 in rust AND ruby; full-corpus sweep 11,016 pass / 7 pre-existing)

• bgnpcgn-tha-Thai-Latn-2002 (6 failures fixed, 12/12): leading vowel glyphs (ไ ใ โ เ แ) now romanize in phonetic order per the source's vowel table + note 1 ("misleading order... as pronounced"), via 344 enumerated digraph rows; plus the note-4 silent-ห rule.
• din-fas-Arab-Latn-31635-2011 (47/47): row 8 renders hamza ʾ wherever present (the word-initial suppression removed; row-1 alif omission kept); the row-15 he rules were encoded with flipped boundary semantics (word-final -> a), fixing gufta/masʾla class; one test corrected to the letter-level masʾla per the هم->hm precedent.
• alalc-khm-Khmr-Latn-2013 (10/10): bantac shortens the PRECEDING vowel per the page-4 notes (35-row strip block + retargeted bantac rows); two test inputs had the u-vowel codepoint typo'd.
• alalc-lao-Laoo-Latn-2012 (10/10): word-initial ອ->ʻ per Note 6; the ຫຍ້າ test input had ໍ (niggahit) typo'd for ້ (tone) - fixed to the Note 5 example.
• alalc-mya-Mymr-Latn-2011 (6/6): ဝ="va" and ွ medial="v" per the 2011 table (77 w-rows corrected); bare-medial rows added per the page-2 examples; test inputs corrected to the source's own examples.

Verification

• rust iscrs: 85/85 for the five maps; full-corpus sweep 11,016 passed / 7 failed (the 7 are batch A's maps, identical on origin/main - no regressions).
• Ruby interpreter (.wt-rb-verify bypass): 85/85 parity.