[Feature] Automated benchmark suite for copilot evaluation

[pskeshu]

Description

Create a benchmark task suite to automatically evaluate copilot success/failure on standard microscopy workflows.

Requirements

1. Benchmark Categories

• Navigation: "Move to embryo X", "Center the brightest embryo"
• Acquisition: "Acquire a volume", "Start timelapse for 1 hour"
• Analysis: "Find the hatching embryo", "Measure embryo sizes"
• Multi-step: "Calibrate all embryos and start timelapse"
• Error Recovery: "Handle stage limit error", "Recover from failed detection"

2. Evaluation Metrics

• Task completion (binary + partial credit)
• Tool call efficiency (optimal vs actual)
• Time to completion
• Error handling quality

3. Mock Hardware

• Simulated responses for all device operations
• Configurable failure scenarios
• Deterministic for reproducibility

4. Reporting

• Per-task results
• Aggregate scores
• Regression tracking over time

Technical Approach

• Define benchmark tasks in YAML/JSON format
• Create MockQueueServerClient with scripted responses
• Run copilot in evaluation mode (no human input)
• Score based on tool calls and final state

Key Files

• New: benchmarks/tasks/
• New: benchmarks/mock_client.py
• New: benchmarks/evaluator.py