Make subagents not use additional requests when the provider supports it

[lake-uproar]

What are you proposing?

make zed subagents run or be counted under the parent session so starting a subagent doesn't fire another billable LLM request.

Why does this matter?

I tested the subagent pull requests with github copilot as the backend and subagents each use billed requests even though github copilot supports subagents.

Other LLM providers probably also support subagents but I only have github copilot.

Are there any examples or context?

Subagents in vscode's copilot chat extension are both isolated and also don't use extra requests.

Possible approach

looking at how https://github.com/microsoft/vscode-copilot-chat and https://github.com/github/copilot-sdk work? I'm not sure, i'm not a dev

[amio]

A session spinning up three sub-agents will increase Copilot's Opus pricing ratio to (1+3)*3=12x 😲