Add power consumption tracking and budget constraints for owned/local GPUs

[anfredette]

When users deploy on owned or local GPUs (rather than cloud-rented), power consumption becomes a significant operational cost and constraint that isn't captured in current pricing models.

Requirements

1. Power budget constraint - Allow users to specify a power budget (in watts or kW) that the deployment must stay within. This could be:

   • A hard limit (reject configurations exceeding budget)
   • A soft preference (penalize configurations in scoring)

2. Power-aware cost calculation - Include electricity costs in the total cost of ownership:

   • GPU TDP (Thermal Design Power) data for each GPU type
   • User-provided electricity rate ($/kWh)
   • Utilization-adjusted power draw estimates
   • Monthly power cost = GPU count × TDP × utilization × hours × rate

Additional considerations

• Cooling overhead - Data center cooling typically adds 30-50% to raw power consumption (PUE factor)
• Multi-GPU power scaling - Power draw doesn't scale linearly with GPU count due to interconnect overhead
• Idle vs active power - GPUs draw less power when idle; consider expected utilization patterns
• Power efficiency scoring - Add power efficiency as a potential ranking dimension (tokens/watt)

Data needed

• GPU TDP values for each supported GPU type (already partially available in hardware profiles)
• Typical utilization-adjusted power consumption benchmarks