Add runner abstraction for multi-tool support

[that-github-user]

Summary

• Define Runner interface: name, description, available(), run()
• Runner registry with getRunner(), listRunners(), getDefaultRunner()
• Refactor Claude Code runner to implement the interface
• Add --runner CLI flag (default: claude-code)
• Tool availability check before running (clear error if not installed)
• Foundation for adding Aider, Cursor, Copilot runners

Change type

• Bug fix
• New feature
• Refactor
• Documentation
• CI / infrastructure
• Chore

Related issue

Closes #32

How to test

npm test                    # 61 tests pass
thinktank run --help        # shows --runner flag
thinktank run "task" --runner nonexistent  # "Unknown runner" error

Breaking changes

• This PR introduces breaking changes

The runClaudeAgent export is replaced by claudeCodeRunner object export. No external consumers expected at this stage.

🤖 Generated with Claude Code

[that-github-user]

Self-review checklist:

• Runner interface is minimal: name, description, available(), run()
• Registry returns undefined (not throws) for unknown runners
• run.ts checks availability before creating worktrees
• Error messages are actionable ("Is X installed?")
• Claude Code runner preserves all existing behavior (spawn, timeout, diff capture)
• --runner flag defaults to claude-code, visible in --help
• 5 new registry tests, 61 total passing
• Lint clean, types pass