Tweak/disable thinking for GPT models and inline chat

[jrieken]

The thinking configuration we used for our endpoints is mostly defined by the agent-use-case. Other usages, like for inline chat or via the API, inherit these configurations and we should rethink that

[roblourens]

First, there's a complex matrix for which models support which reasoning effort levels. Some support a none but not all. Some have minimal and low, some only have low, and some have xhigh of course. But using none or minimal is "not recommended for coding use cases". So I wonder if you'd like to use "low" for inline chat.

I tried one simple testcase of asking inline chat to add a doc comment with gpt-5.2. With "low", the requests averaged 5.5s. With "none", they averaged 4.5s, but consistently used // instead of the /** I was looking for.

Try it with github.copilot.chat.responsesApiReasoningEffort if you want. It will use "none" if you set that, don't mind the warning.

Second, I wonder if you have an opinion about the default API value. I think that the "medium" we use for the agent is a good default.

[roblourens]

Will finish for Jan microsoft/vscode-copilot-chat#2720