Support copilot `--resume "<query>"` for natural-language session lookup

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

Support using `copilot --resume` with a natural-language query to find a past session when I do not know the session ID or when session names are vague, duplicated, or not meaningful. Today, if multiple sessions begin with similar prompts or were never renamed, `copilot --resume` is not very helpful for finding the right conversation. My current workaround is to keep a separate "search-history" session and ask Copilot things like "find me the conversation about PR # 1234". That works, but it is indirect and inconvenient.

### Proposed solution

`copilot --resume "conversation about PR #id"`

Output is a list:
sessionId1 | Short description | confidence %
sessionId2 | Short description | confidence %

### Example prompts or workflows

- `copilot --resume "conversation about PR #1234"`
- `copilot --resume "session where we debugged the auth token refresh bug"`
- `copilot --resume "discussion about payments API refactor"`
- `copilot --resume "the conversation where we edited src/auth/login.ts"`
- `copilot --resume "session about the failing CI job on Windows"`

### Additional context

If I was talking about issue #id in the context of design in one conversation, in another I was developing the fix for the same issue and in the 3rd I was reviewing an old PR submitted as a fix to the same issue, I might have multiple conversations starting with same prompt "fetch issue #id" around the same time and using `copilot --resume" doesn't help me enough. I would have to open each session and read through the messages or name them meaningfully.

Might be related: https://github.com/github/copilot-cli/issues/714

Start at the existing `copilot --resume` entry point and review the related issue #714 for context. Define how a natural-language query should find and rank past sessions, then verify that the command lists session IDs, short descriptions, and confidence percentages for the example workflows.

由索引模型根据 Issue 内容生成。