fix: Add PermissionRequest hook workaround for piped command permissions (#11775)

[AnthonyMDev]

Summary

• Adds a PermissionRequest hook example (examples/hooks/piped_command_permission_fix.py) that works around the regression where the Plan agent prompts for permission on piped commands even when both individual commands are whitelisted in settings.json
• The hook intercepts Bash permission requests, checks if the command is a simple pipeline where each segment's base command is in a configurable allowed list, and auto-approves if all segments match
• Includes safety checks to reject commands with dangerous patterns (&&, ||, $(, backticks, ;, <<)

Root Cause Analysis

The regression was introduced in v2.1.7's security fix ("Fixed security vulnerability where wildcard permission rules could match compound commands containing shell operators"). In the permission checking flow:

1. FXq (pipe handler) correctly splits the piped command, validates each segment individually, and returns "allow"
2. After FXq returns, jF (bash security checker) runs on the original compound command and flags the pipe operator
3. This overrides the "allow" result back to "ask", causing the permission prompt

The proper source-level fix would be to skip the jF security check when FXq has already validated all pipe segments individually (i.e., when decisionReason.type === "subcommandResults"), since the security of each segment has already been verified.

Test plan

• Install the hook in .claude/settings.json per the docstring instructions
• Configure ALLOWED_PREFIXES to match your permissions.allow rules
• Run a piped command like ls -lh *.webp | awk '{print $5, $9}' with both ls and awk whitelisted
• Verify the Plan agent no longer prompts for permission on the piped command
• Verify that piped commands with non-whitelisted commands still prompt for permission
• Verify that commands with dangerous patterns (&&, ||, etc.) are not auto-approved

Fixes #11775