fix: preserve custom LaTeX env shorthand macros during translation

[octo-patch]

Fixes #2240

Problem

Physics and math papers frequently define shorthand commands for equation environments using \def or \newcommand, e.g.:

\def\be{\begin{equation}}
\def\ee{\end{equation}}
\def\bea{\begin{eqnarray}}
\def\eea{\end{eqnarray}}

The LaTeX segmentation logic in latex_actions.py only recognized standard \begin{equation}...\end{equation} patterns. When authors used these shorthands, the enclosed math content was not preserved from GPT translation — causing it to be sent for translation (corrupting the math) or replaced with incorrect LaTeX delimiters like \[.

Solution

Added extract_user_defined_env_patterns() in latex_toolbox.py that:

1. Scans the merged TeX content for \def\cmd{\begin{envname}} and \def\cmd{\end{envname}} declarations (also handles \newcommand / \renewcommand)
2. Finds matching begin/end pairs for the same environment name
3. Returns regex patterns for each matched pair with a negative lookahead (?![a-zA-Z]) to avoid false matches on longer commands that share the same prefix (e.g., \be must not match \begin)

These patterns are then applied in split_subprocess() immediately after the standard equation/align patterns, so the custom shorthand blocks are correctly preserved from GPT translation.

Testing

Manually tested with a synthetic TeX document containing \def\be{\begin{equation}} and \def\ee{\end{equation}}:

• \be...\ee blocks are correctly marked as PRESERVE
• \begin{equation} is NOT accidentally matched by the \be pattern (negative lookahead works)
• No changes to existing behavior when no custom shorthands are defined