Need a way to programmatically start a session/subagent with specific context window and reasoning effort

### Describe the feature or problem you'd like to solve

Currently model's context window size and reasoning effort can only be manually specified through the interactive `/model` flow. There seems to be no way to specify it for subagent and SDK.

### Proposed solution

- Agent markdown file's frontmatter adds some properties to specify the reasoning effort and context window level.
- Copilot SDK exposes a supported way to specify reasoning effort and context window tier when creating a session.

### Example prompts or workflows

- Although some models have larger context window (like 1M), but by default they use a smaller one (like 200K), which limits their performance.
- For automation system, it needs a more powerful model with more reasoning effort and larger context window to do complex tasks for large project.
- Multi-agent workflow requires delegating tasks to subagent, and may need different subagent to have different settings.

### Additional context

_No response_

Start by locating the agent markdown frontmatter handling and the Copilot SDK session-creation API. Compare how the interactive /model flow sets reasoning effort and context window, then trace how subagents are started. Done means agents, SDK sessions, and subagents can each receive these settings programmatically.

由索引模型根據 Issue 內容生成。