fix: #3579 conditional parsed as optional chaining

[josdejong]

Fixes #3579

Solves math.evaluate('true?.3:.7') from being parsed as optional chaining and throwing an error.

[gwhitney]

Hmm, it's very special-case. I worry just slightly about the tokenizer becoming cluttered with lots of such special cases over time. This could be merged, but is it worth taking a second to consider whether any of the following would be cleaner:

• Do this "checking for a number before tokenizing . as part of a delimiter" for all delimiters? Note that could regularize the parsing of the single-character . delimiter which is already so special that ., somewhat confusingly, isn't even included on the list of delimiters?
• Allow delimiters to be regexps, so they can include negative lookahead assertions? That would eliminate the need to separately loop over 3, 2, and 1-character delimiters, and should also allow unifying the . single character delimiter?
• Handle delimiters as a tree of characters rather than a list of 1, 2, and 3-character strings, so that as long as you are within a valid delimiter, you keep accumulating characters, except you never accumulate a . that is followed by a digit. That should also allow the handling of single-character . to be unified with all of the other delimiters.

Just some thoughts. If you prefer, I am willing to take a quick stab at any of the above, whichever sounds best to you -- none of them should take long, and I think any of them would overall simplify the tokenization rather that just adding another wrinkle to it. Let me know. Or I can just merge this as is if you prefer.