手工导入的模型缺少窗口、视觉和推理配置

来自 #4402 的 P2：连接编辑只改连接，模型能力应在统一模型设置里配。当前发现只返回 ID、或用户手工添加的模型只落 `id`/`name`，统一设置里的上下文项写的是任务工作上限，补不了 `contextWindow` / `supportsImageInput` / `reasoning`。这些型号会卡在默认 200K，也无法打开视觉或推理。

本 PR 不再把能力编辑塞回连接弹窗。后续在标准模型设置里补这些字段。

The issue identifies manually added models and the standard model settings as the relevant areas, with contextWindow, supportsImageInput, and reasoning as the missing fields. Start by tracing how manually added model records and unified settings are represented, using #4402 as context. Done means these settings supply all three capabilities without putting them in the connection editor.

Written by the indexing model from the issue text.