feat: LocalModel skill — privacy-preserving local LLM classifier

[N4M3Z]

Summary

• New LocalModel skill for routing prompts to a local LLM when privacy demands it instead of sending content to a cloud model
• classify.py helper: classifies a prompt against a label set using the local backend
• detect-backend.sh helper: probes for a usable local LLM backend (ollama, llama.cpp server, etc.) and prints the endpoint

Split out of #49 so it ships as a standalone change.

Test plan

• forge validate passes locally
• CI green on the PR
• Spot-check forge install --target ~ --force deploys the skill cleanly
• Both helper scripts execute (chmod +x preserved through cherry-pick)