core: remove special execve handling for skill scripts

[bolinfest]

Why

We still want general skill support, but we no longer want executables under a skill's scripts/ directory to get bespoke exec approval or escalation behavior.

With this change, intercepted skill scripts follow the same path as other commands, so the turn sandbox and normal exec policy stay in charge. Declared skill permissions in openai.yaml no longer widen execution for scripts/ helpers.

What Changed

• removed the SkillScript branch, find_skill() lookup, and skill-specific approval wiring from core/src/tools/runtimes/shell/unix_escalation.rs
• removed the execve session approval cache that only existed for the skill-script path
• kept general skill support intact; SKILL.md loading and openai.yaml parsing are unchanged
• replaced the old integration coverage with focused tests that assert declared skill permissions are ignored for script execution and that WorkspaceWrite still blocks out-of-workspace writes under zsh-fork