texmath: infinite loop on block math opened in a blockquote and closed after it

[acaprino]

Describe the bug

With texmath_plugin, a display-math block that opens inside a blockquote and whose closing delimiter only appears after the blockquote makes md.parse() loop forever, with memory growing until the process is killed.

Minimal reproduction

from markdown_it import MarkdownIt
from mdit_py_plugins.texmath import texmath_plugin

md = MarkdownIt("commonmark").use(texmath_plugin)  # delimiters="dollars"
md.parse("> $$ a=1\n\n$$\n")  # never returns

Every delimiter set is affected:

delimiters	input
dollars, kramdown	"> $$ a=1\n\n$$\n"
brackets	"> \[ a=1\n\n\]\n"
gitlab, julia	"> ```math a=1\n\n```\n"

Nested containers too ("> > $$ a=1\n\n$$\n", "- > $$ a=1\n\n$$\n"). Lists on their own do not hang.

Reproduced with mdit-py-plugins 0.6.1 and master (519953d), markdown-it-py 4.2.0 and master (6f58654), Python 3.12, Windows 11.

Cause

make_block_func in mdit_py_plugins/texmath/index.py matches the rule's regex against state.src[begin:], the rest of the whole source, so the match can end past endLine. The loop that looks for the closing line then finds nothing and state.line is never advanced, but the rule still pushes a token and returns True. The block tokenizer calls it again on the same line, forever, pushing a new token each time.

It is the same failure mode that #117 fixed for amsmath in 0.4.2. It is not covered by #148, which fixes regex backtracking.

Suggested fix

Look for the closing line only within [begLine, endLine), and return False before pushing any token when the match ends outside the current block: no auto-closing, as in amsmath. A PR with this change and regression tests follows.

markdown-it JS turns this situation into an error since 13.0.2 (block rule didn't increment state.line, markdown-it/markdown-it#847); markdown-it-py has no such guard, so I'm opening a separate issue there.