Describe the bug
tools/fst/prepare_dict.py, it can't split chinese phrase into character when bpemodel given.

To Reproduce
tools/fst/prepare_dict.py lang_char.txt lexicon_raw.txt lexicon.txt bpe.model

if lexicon_raw.txt have a phrase "我们":

sp.EncodeAsPieces("我们")
['▁', '我们']

Expected behavior
it will prompt contains oov unit “我们”， we want it be ['我', '们'] actually.

Screenshots

Additional context
One of the ways to modify it :
if not word.encode('UTF-8').isalpha():
pieces = list(word)
else:
pieces = sp.EncodeAsPieces(word)
We assume that one entry of lexicon_raw.txt only one language.