Teaching INGRID to reason about heat

Until this milestone, BlueCrow could tell you a temperature was high. It couldn't tell you whether that was expected.

That's the gap the new Thermal Intelligence module closes. It looks at CPU and GPU load and temperature together, not in isolation, and answers two questions a raw number can't:

• Is this workload CPU-bound, GPU-bound, or balanced? Based on which component is actually under load right now.
• Does an elevated reading look explained by that load, or not? A hot GPU during a heavy render is a different situation than a hot GPU sitting at 5% usage.

A few things we were deliberate about:

• Facts stay separate from inference. Every finding says plainly what was measured versus what's being inferred from it, and the module never invents a value it doesn't have.
• Severity never gets minimized by "well, it's under load." An explained high temperature is still a high temperature. Load context changes the narrative and confidence, never the verdict.
• It builds on the same shared thresholds the dashboard's health score already uses (70°C / 85°C / 95°C), rather than inventing a second, competing scale.

This is the first BlueCrow reasoning module to synthesize across two live metrics instead of reporting on one. Storage and memory reasoning are next on the roadmap, following the same shape.