fix: expose max thinking level for reasoning models

[damngamerz]

Summary

Expose the max thinking level for Requesty models in pi.

Right now, pi-requesty registers discovered models with reasoning: true but no thinkingLevelMap, so pi only shows min / low / med / high — never max — even for models that support it.

Fix

Map each reasoning-capable model with thinkingLevelMap: { max: "max" } (see discoverModels), and persist thinkingLevelMap in updateModelsJson so /requesty-models-sync keeps it.

Why this is safe for every reasoning model

Requesty's /models response carries no machine-readable level field (only supports_reasoning). However, max is a global Requesty-reasoning-effort value that Requesty normalizes per backend (see https://docs.requesty.ai/features/reasoning):

• OpenAI → forwards "high"
• Anthropic → converts to a token budget near max output tokens
• Vertex / Gemini → converts to a token budget near max output tokens

It is never rejected for any model type (only xhigh is model-specific). pi sends the mapped value literally as reasoning_effort, so Requesty performs the normalization.

Changes

• requesty.js: inline thinkingLevelMap: reasoning ? { max: "max" } : undefined in the discovery mapping; persist it in updateModelsJson.

Fixes #8