fix: crash on inputs containing runs of 4 spaces

[lukors]

fixes: #7

The processors register their replace_tokens values with the tokenizer via tokenizer.add_tokens(). The 4-space string used for the "\t" -> spaces substitution seems like it has no row in the sembr2023 model's embedding matrix. This means any input with four or more consecutive spaces therefore produces an out-of-range token ID and crashes with:

IndexError: index out of range in self

Reproducer:

printf 'hello    world' | uvx sembr

The fix in this PR is to not register the 4-space string with the tokenizer. Since its registration was a result of being a value in replace_tokens, I removed it from the dict and handle the tab-to-spaces substitution separately:

• MarkdownProcessor, LaTeXProcessor, PlainTextProcessor: drop the "\t": " " * self.spaces entry from _get_replace_tokens.
• MarkdownProcessor.parse_text: add an explicit text.replace("\t", " " * self.spaces). The other two processors already did this substitution.
• BaseProcessor: drop the now-dead "if k != '\t'" filter from reverse_replace_tokens.

[admk]

Thanks, merged. I have not tested it though, hope it fixes the bug and doesn't have regressions. I hope to plan to add tests later...