fix(chain-g2p): convert non-eng dicts to tab-separated so DictStep loads them

The six ChainG2p-* *_dict.txt files separate word and phonemes with two (or one) spaces, but PhonemeDict::load only accepts tab-separated lines and silently skips every line without a tab. The dictionaries therefore always loaded EMPTY, every lookup fell to the multig2p model, and the two-pass dict never actually resolved words on these packages (the L2 'non-eng chains hit the dict' test passed, but only because the model coincidentally produced the same phonemes; НЕТ was the counter-example, giving 'y y' from the model instead of the dict's 'nn je t'). Convert the first whitespace run to a tab (verified: only the separator changed, no other corruption). Kor was already tab-separated; Fil is excluded (uppercase dict, no cleaner).

Author: wolfgitpr