fix: prevent main model thinking variant from applying to small model - #6839

Summary

When changing thinking level (ctrl+t) for the main model, the selected variant (e.g., reasoningEffort) was incorrectly applied to small model calls (title/summary generation). This caused API errors when the small model didn't support the parameter.

Problem

small=true agent=title params={"options":{"reasoningEffort":"low"}}
ERROR "Model grok-code does not support parameter reasoningEffort."

The variant from input.user.variant was applied regardless of whether small: true was set.

Fix

Skip variant application when small: true in llm.ts:

// Before
const variant = input.model.variants && input.user.variant ? input.model.variants[input.user.variant] : {}
// After
const variant = !input.small && input.model.variants && input.user.variant ? input.model.variants[input.user.variant] : {}