fix invalid_characters bug

[ain-soph]

There are 2 issues.

1. In Chinese/English dataset texts, there are often "?" or "？". Currently, they are regarded as invalid characters but no map in character_simplier.
2. We should call count_invalid_characters again after _apply_character_map. Otherwise, if there is still no tag in text, It would go into the next if condition and raise an error because tags are None.

ChatTTS/ChatTTS/norm.py

Lines 194 to 200 in 69aa900

	if len(invalid_characters):
	texts, tags = _split_tags(text)
	self.logger.debug("split texts %s, tags %s", str(texts), str(tags))
	texts = [self.reject_pattern.sub("", t) for t in texts]
	self.logger.debug("normed texts %s", str(texts))
	text = _combine_tags(texts, tags) if len(tags) > 0 else texts[0]
	self.logger.debug("combined text %s", text)

[ain-soph]

This PR fails #655 unittest, which requires a further check.

[ain-soph]

The first change in character_simplier seems not to be the expected behavior according to #655 test file, as it requires the refined text still has ? inside.

I wonder since this ? is invalid_character, which means it's not seen in training data. Shall we still keep that ? in inference stage?

[ain-soph]

After check, it seems both concerns are not expected behaviors. Close this issue.