[BUG] Reasoning-capable models default to no thinking after model selection

[zoomote]

Description

Optional-reasoning models can resolve to disable after selecting or switching models even when their model metadata advertises a reasoning default. This leaves models such as OpenCode Go DeepSeek V4 / GLM 5.2 and NanoGPT Muse Spark running without visible thinking unless the user manually reselects a reasoning effort.

Originally reported in Discord: https://discord.com/channels/1497384592494297201/1504099932158824590/1540812344861663393

Reproduction

1. Use Zoo Code v3.80.0.
2. Select OpenCode Go and deepseek-v4-flash with no persisted reasoningEffort.
3. Inspect the reasoning-effort selection.

A focused component reproduction renders the selected effort as disable, although the model metadata declares reasoningEffort: "high". The same shared settings path affects other optional-reasoning models.

Expected behavior

When no explicit user choice is persisted, a reasoning-capable model should use its advertised default effort and enable reasoning. An explicit user selection of disable must continue to turn reasoning off.

Actual behavior

The settings UI defaults optional-reasoning models to disable. Model switching clears reasoningEffort, and NanoGPT only forwards an explicitly stored effort, so affected models may not think.

Technical notes

• ThinkingBudget chooses disable as the default unless reasoning is required, ignoring an optional model default.
• Model-change side effects clear reasoningEffort but do not establish the newly selected model default.
• NanoGPT request construction reads only the persisted effort rather than falling back to model metadata.

Acceptance criteria

• Add regression coverage first for an optional-reasoning model with an advertised default and no stored effort.
• Default to the model-advertised effort when the user has not made an explicit selection.
• Preserve explicit disable behavior.
• Cover request construction for affected providers where needed.