Bash parser adds zsh arithmetic guard and tracks variable assignments

What

Before Claude Code runs a shell command, it parses the command (breaks it into its parts) to decide whether it is safe or needs your permission. That parser changes in this release:

• zshMathGuard: when it is set, the parser handles # followed by $, and flags a (( sequence as zshMathDiff. This is about zsh arithmetic, which zsh can read differently from bash.
• Variable assignments: the parser now collects nested variable_assignment parts of a command into an assignments list.
• for loops: the parser tracks the head of a for loop. What happens when parsing fails for $(( )) and for (( )) also changes.

Separately, compaction (summarizing a conversation to free up space) gains handOverAtStop when keepThroughStop is set.

Why

These parts feed the Bash permission check, so commands that use zsh arithmetic or variable assignments may be judged differently than before.