feat: Code_Puppy destructive command authZ engine

[collin-kierce]

This PR aims to improve the way destructive commands are handled by code-puppy. This is meant to help keep users from accidentally running malicious commands on their devices.
What is currently in place:

• A cheap substring search looking for key commands such as "rm", "format", "rmdir"
• A "is_real_invocation" function which checks to see if the found keyword is the first word in the command
• An expensive regex matching to confirm wether this is a malicious command
• If any of these three stages don't find anything suspicious the command will run fine. If all three stages agree it's a harmful command, the user will be alerted about the harmfulness of the command and prompted for permission to proceed or stop.
• Two different callbacks/guardrails for destructive commands and git commands
• All data regarding commands and substrings is stored inside file
• iterates through all expensive regex patterns looking for a match

What this PR does:

• Start by normalizing command, removes common obfuscation methods such as empty "" and ''
• Splits command on compound operators such as && or || to check all parts of compound command
• Conducts a cheap search for keywords in the command/commands
• Does not do a "is_real_invocation" check as this was leading to false negatives with commands such as sudo rm -rf /
• Matches regex patterns to see if there is a hit or not
• If there is a regex match the command will either be blocked immediately or the user will be prompted for permission depending on what the command is. A command like rm -rf / is blocked immediately
• Merges Git guardrail and destructive command guardrail into one
• Loads all data from JSON files inside new directory called patterns
• Separates expensive regex patterns by categories and only searches through categories found in cheap substring search