/agent <name> - Allow direct invocation of agents by name

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

Currently, `/agent` only opens an interactive browser/picker with no way to pass an agent name as an argument. Users who know the agent name they want must scroll through the full list every time. This adds friction, especially with many agents installed.

### Proposed solution

Support /agent to directly invoke a named agent (e.g., `/agent my-agent`). If the name doesn't match, fall back to the interactive picker filtered by the input string. This mirrors how /model and other commands accept optional arguments.

### Example prompts or workflows

1. Quick invoke by name: `/agent my-test-agent` - immediately starts the agent without navigating the picker. Saves time when you know what you want.
2. Fuzzy match with fallback: `/agent test` - filters the agent list to matches containing "test", showing only relevant options. Falls back to filtered picker when no exact match.
3. Chained with context: `/agent my-test-agent generate tests for src/auth/handler.ts` - invoke agent and pass initial prompt in one command, skipping the extra round-trip.
4. Scripting/repeatability: User frequently runs the same agent on different files. `/agent my-agent` lets them hit up-arrow, change the file path, and re-run - not possible with an interactive picker.
5. Discoverability without friction: `/agent --list` to quickly print available agent names (one-line each) without entering the interactive picker, so users can learn names then invoke directly.

### Additional context

_No response_

Start at the /agent command entry point and compare its argument handling with /model and other commands. Exercise exact-name invocation, filtered fallback, chained prompt text, and --list; done means these workflows work without breaking the existing interactive picker.

Written by the indexing model from the issue text.