feat: add math-delimiter validator - #22

Some notes from reviewing this before merge. I ran the checker from this branch (Python 3.11) on a few short samples rather than reading it alone. Nothing here is a blocker.

A $ inside code is reported as a maths error. Both of these come back as INVALID_NEWLINE_INSIDE_INLINE:

from in2lambda.validation import math_delimiter_checker as chk
chk("Run `echo $PATH` now.\n")      # INVALID_NEWLINE_INSIDE_INLINE
chk("```bash\necho $HOME\n```\n")   # INVALID_NEWLINE_INSIDE_INLINE

Any question that shows a shell command or a code sample gets a warning it shouldn't. Skipping code spans and fenced blocks before scanning would fix it. (Prices are fine, by the way: It costs $5 and then $10. passes.)

The message for an unclosed $ is misleading. chk("This $x is unclosed.\n") also returns INVALID_NEWLINE_INSIDE_INLINE, where MISSING_CLOSING_SINGLE_DOLLAR is what the author needs to read.

No location, and only the first problem. The checker returns one enum for the whole document, so on a file of twenty questions you learn that something is wrong somewhere. Returning a list of messages with line numbers would make it actionable, and would let the caller print all of them. Tracking the line is only a few lines of code given the loop already walks the string.

One documentation point. delimiters.py says KaTeX requires $$ on its own line. As far as I know that's our house convention rather than a KaTeX rule, so it's worth rewording to say so.

Following up on my last point above, to correct it before anyone acts on it.

I said the $$-on-its-own-line rule was "our house convention rather than a KaTeX rule". That's too weak, and it could be read as an argument for relaxing the check. It isn't one — the rule should stay exactly as it is.

in2lambda/filters/markdown.py:169 already emits display maths as:

f"\n\n$$\n{expression}\n$$\n\n"

with expression having gone through latex_to_katex() first. So the checker is enforcing precisely the form in2lambda itself writes into masterContent and the part/solution fields. I ran the shapes that line produces — inline, display, the two mixed, and a display block at end of file — and they all pass the checker.

What's actually wrong is only the attribution in delimiters.py:3-5. KaTeX never sees delimiters: it takes a maths string plus a display flag, and scanning for $/$$ belongs to the renderer integration in front of it. So "KaTeX expects ..." is the wrong justification for a rule that is right on its own terms.

Suggested reword: say the check matches the output format in2lambda emits (filters/markdown.py:169) and that Lambda Feedback renders, rather than citing KaTeX. The other three points above stand as written.

Updated to better handle $ inside code, by specifically searching for code environments.

Updated error handling to report line number and file.

Made it clearer that $$ on a new line is our convention.