[Feature] Add footprint and throughput measurement

[HectorHHZ]

Tables 5 and 6 compare deployment cost of base and pruned checkpoints.

Scope

• Total and active parameters, disk size, peak VRAM, and throughput (tokens/s) under a fixed vLLM workload.
• The same workload and hardware for base and pruned checkpoints, recorded in the results schema.

Depends on: #5

Done when

• Measurements for base and pruned Qwen3.5-35B-A3B are posted here.

Part of #2