Add run_scenarios tool to MCP server for batch testing

[richardkiene]

Summary

Adds a new run_scenarios tool to the MCP server that accepts a list of scenario paths and runs them all sequentially, returning an aggregated summary.

Problem

Previously, Claude Code had to call run_scenario one at a time, making batch testing inefficient.

Solution

New run_scenarios tool that:

• Accepts a list of scenario paths
• Runs each scenario sequentially
• Returns aggregated summary with pass/fail counts

Example output:

## Test Run Summary
**2/3 passed** (1 failed)

### Results
- ✓ `scenario1.yaml`: PASSED (score: 0.95)
- ✓ `scenario2.yaml`: PASSED (score: 0.88)
- ✗ `scenario3.yaml`: FAILED (score: 0.42)

Changes

• Add ScenarioRunResult dataclass for structured results
• Extract _execute_scenario helper for reuse
• Add run_scenarios tool with summary formatting
• Add tests for new tool

Test plan

• All 258 unit tests pass
• Ruff linting passes
• Mypy type checking passes