[codex] support hook input rewrites

[abhinav-oai]

Why

Hooks can currently inspect tool input, but callers cannot safely substitute a rewritten payload while preserving the approval semantics around that new payload. That leaves sanitizing or normalizing hook workflows unable to change what actually runs.

What

• Add updatedInput support for PreToolUse and PermissionRequest hook outputs.
• Let PreToolUse rewrites carry allow or ask semantics into shell, apply_patch, unified exec, and MCP execution paths.
• Let PermissionRequest rewrites replace the full input, then re-enter normal approval evaluation so rewritten input is checked again before execution.
• Regenerate hook schemas and add coverage for shell/apply_patch/MCP rewrites, ask parsing, no-op handling, and rewritten-input re-evaluation.

Validation

• cargo test -p codex-hooks
• cargo test -p codex-core --lib mcp_tool_call::tests::permission_request_hook_can_update_mcp_tool_input
• cargo test -p codex-core --test all suite::hooks::permission_request_hook_rewrites_shell_command_before_execution -- --exact
• cargo test -p codex-core --test all suite::hooks::permission_request_hook_rechecks_rewritten_shell_command -- --exact
• cargo test -p codex-core --test all suite::hooks::pre_tool_use_rewrites_shell_command_before_execution -- --exact
• just fmt
• just fix -p codex-core
• just fix -p codex-hooks

cargo test -p codex-core --lib was also attempted; it aborted in the existing tools::handlers::multi_agents::tests::multi_agent_v2_close_agent_accepts_task_name_target test with a stack overflow unrelated to this change.