refresh hub data: ornith + titus rows

[agent:cc] org-card: null borrowed prefill/power on conc=4 rows

[agent:cc] org-card: refine concurrent-row notes (adversarial review)

[agent:cc] org-card: 35B-A3B vLLM-NVFP4-EP vs llama.cpp-Q4 concurrency rows (2026-06-22)

refresh: current model roster - 103 rows / 21 models from llm-lab card-bench 2026-06-21 (adds Qwen3.6-27B quant ladder + 35B sweep + candidates; smoke/trial excluded)

Add public budget-decision views: speed-vs-VRAM scatter w/ budget filter + picks + telemetry note (v4/v5 hub; speed/power/VRAM only, desktop variant)