feat(opencode): add optional Jev auto-approve backend - #2

I built an optional TypeSafe Jev backend on top of this branch, without changing the default classifier.

It is one commit on ac6133f633b2b5bed0a9bebf45e21041455fa065. Setting auto_approve.backend to jev classifies each consequential action as approve, ask, or deny: approve runs it, ask opens the normal prompt, and deny rejects it.

TypeSafe is not in the built-in provider list, so the key goes through the existing custom-provider login:

1. Run opencode auth login.
2. At Select provider, choose Other.
3. At Enter provider id, enter typesafe. The login warns that this only stores a credential and that the provider still needs configuration in opencode.json.
4. At Enter your API key, paste the TypeSafe key. The prompt hides the input and writes it to OpenCode's auth store. It does not belong in opencode.json.

Then enable the backend:

{
  "$schema": "https://opencode.ai/config.json",
  "experimental": { "auto_approve": true },
  "auto_approve": {
    "backend": "jev",
    "jev": { "model": "jev-latest" }
  }
}

jev.model is optional and defaults to jev-latest. The question labels stay fixed, but the guidance and acceptance threshold are configurable. Omit any field to keep its built-in wording. min_probability defaults to 0.95.

"jev": {
  "model": "jev-latest",
  "instructions": "Your trusted policy guidance.",
  "criteria": {
    "approve": "When to run it.",
    "ask": "When to prompt.",
    "deny": "When to reject it."
  },
  "min_probability": 0.95
}

Restart OpenCode, then press Tab from Build until Auto-approve is selected.