[FEATURE]: Plugin-safe raw text generation

Feature hasn't been suggested before.

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

Describe the enhancement you want to request

Plugins that need a small isolated model call currently have to use client.session.prompt(). That creates a session and runs the normal agent pipeline, adding base prompts, project instructions, tools and context. A plugin-safe raw generation API should reuse OpenCode's provider authentication and model configuration while sending only an explicit system prompt and user prompt.

Suggested contract:

await client.experimental.generateText({
  query: { directory },
  body: {
    model: { providerID, modelID, variant },
    system: "You are a professional translator. Output only the translation.",
    prompt: "Translate this text...",
    temperature: 0,
    maxOutputTokens: 2048,
  },
})

Requirements:

• Reuse existing provider resolution, API keys, OAuth, endpoints, variants, retries and timeouts.
• Exclude session history, agent prompts, project instructions, tools, skills, MCP context and plugin system transforms unless explicitly requested.
• Do not create or persist sessions/messages.
• Do not expose credential material to plugins.
• Return generated text, usage and typed provider errors.

This would let translation and similar plugins make a minimal model request without requiring duplicate credentials or paying for unrelated OpenCode context.