Show context window size (tokens) in /model command alongside premium request multiplier

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

The `/model` command shows the premium request multiplier for each model but not the context window size (max tokens). There is no way to know, from within the CLI, how many tokens a model actually supports in Copilot.

### Proposed solution

Add the context window size (in tokens) to the output of `/model`, alongside the existing premium request multiplier.

**Current `/model` output:**
```
claude-sonnet-4.6 ×1 (default)
claude-opus-4.6 ×3
```

**Proposed `/model` output:**
```
claude-sonnet-4.6 ×1 160k ctx (default)
claude-opus-4.6 ×3 160k ctx
```

The context window values should reflect what **Copilot actually makes available** (i.e. the `max_prompt_tokens` configured in its routing layer), not just the provider's theoretical max. This distinction matters because the effective context window Copilot exposes may differ from the provider's advertised limits — and today users have no visibility into this.

### Example prompts or workflows

1. **Choosing between models for a long session**: A developer about to start a large refactor touching 30+ files runs `/model` to pick the model with the largest effective context window, not just the cheapest in premium requests.

2. **Debugging degraded responses**: A user notices the model is repeating tool calls and re-reading already-processed files. Knowing they're at 160k/200k tokens (vs just "78%") helps correlate the degradation with the known "lost in the middle" effect at high context fill.

3. **Cross-client comparison**: A user switching between VSCode Copilot, Copilot CLI, and Claude Code wants to know if they get the same effective context window across clients. Today this is completely opaque.

### Additional context

- The data is likely already available in the API response from `api.githubcopilot.com/models` (the `capabilities.limits` field), so this should be low-effort to surface.
- Related issues: [#1688](https://github.com/github/copilot-cli/issues/1688) and [#1761](https://github.com/github/copilot-cli/issues/1761) both request configurable compaction thresholds — knowing the absolute context window size would make those features even more useful.