Model reasoning effort missing 'max' (and 'minimal') in Pi integration config

[Smith-106]

Bug: Model reasoning effort missing max (and minimal) in Pi integration config

Environment

• codeg version: 0.23.5
• Platform: Windows 11
• Pi integration: provider config generated with name = "codeg", wire_api = "responses", requires_openai_auth = true

Description

The thinking/reasoning-effort selector in the model settings UI does not offer max, and even when set programmatically, the mapping function rejects it. This limits deep-reasoning workflows on models that support max thinking (e.g. Anthropic/OpenAI/Gemini/OpenRouter providers) when driving Pi.

Evidence

Found in the bundled web frontend (_next/static/chunks/05e5726fe31b6eaf.js):

1. The main settings UI enumeration has no max:

let sy = ["off","minimal","low","medium","high","xhigh"]
sb = [{id:"anthropic",...},{id:"openai",...},{id:"google",...},{id:"openrouter",...}]

2. The mapping function oj() only accepts low/medium/high/xhigh — it rejects both max and minimal:

function oj(e) {
  let t = e.trim().toLowerCase();
  return "low"===t || "medium"===t || "high"===t || "xhigh"===t ? t : null
}

3. Inconsistency: the Kimi (Moonshot) channel enumeration does include max:

let so = ["minimal","low","medium","high","xhigh","max"]  // only used for Kimi

4. Config write path drops the value before it reaches Pi:

model_reasoning_effort = oj(t.modelReasoningEffort) ?? op

Even if max were passed in, oj() returns null so it never reaches the generated Pi provider config. Pi itself fully supports max (thinkingLevelMap maps max → max), so this is purely a codeg-side limitation.

Expected behavior

• The thinking selector should offer max for providers that support it (or at minimum allow max to be passed through), matching the Kimi channel's so enumeration.
• oj() should accept minimal too — it is present in the UI enumeration (sy) but currently mapped to null.

Suggested fix

1. Add "max" to sy: ["off","minimal","low","medium","high","xhigh","max"]
2. Update oj() to accept minimal and max (or align it with the same enumeration used by the UI)