Non-frontier models are unusable without Reasoning Effort selector UI

Currently, GLM/Qwen and other good models reason for literally 10-15 **minutes**, even after i CAPSED them to not do so.

There is a crucial need for any reasoning effort selector UI to make the platform at least usable

No file, test, or entry point is identified in the issue. Start by locating the model-selection UI and how reasoning settings are represented for GLM, Qwen, and other non-frontier models. Done means users can select a reasoning effort level and the chosen setting is applied to supported models without forcing lengthy reasoning.

Written by the indexing model from the issue text.