[FEATURE]: Make bash arity dictionary (BashArity) configurable #46250

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Context: The permission system's "always allow" whitelist pattern is computed by BashArity.prefix(tokens) (packages/opencode/src/permission/arity.ts), using a hardcoded ARITY dictionary keyed by known command prefixes (e.g. git: 2, gh: 3, npm: 2). When a command starts with an unknown wrapper/alias, the lookup fails and prefix() falls back to the first token only.

Concrete problem: For users who wrap shell commands (e.g. the rtk CLI rewrites git push --force origin main into rtk git push --force origin main), the permission request for the rewritten command produces an over-broad "always" whitelist:

• BashArity.prefix(["rtk","git","push","--force","origin","main"]) → falls back to ["rtk"] → "rtk *"
• Clicking Allow always on that request whitelists every rtk ... command, silently defeating per-command permission rules. For example a future rtk gh pr review --comment 123 — which the user's "rtk gh pr review*": "ask" rule was meant to gate — becomes auto-approved, because the approved rtk * rule is matched after the config ruleset via findLast in packages/opencode/src/permission/index.ts.

Request: Make the arity dictionary configurable so users can teach prefix() about wrapper commands/aliases:

1. A config knob to extend/override the built-in ARITY map, e.g. "rtk git": 3, "rtk gh": 4, so prefix() yields rtk git push * / rtk gh pr review *; or
2. A smarter prefix() that skips unknown leading tokens and resolves arity from the first known command, producing the same scoped patterns without config.

Relevant files: packages/opencode/src/permission/arity.ts, packages/opencode/src/tool/shell.ts (line ~409: scan.always.add(BashArity.prefix(tokens).join(" ") + " *")), packages/opencode/src/permission/index.ts (evaluate/findLast + approved rules).

Environment: opencode-ai 1.18.25, Linux