fix: restore per-host probe coverage + sweep residual retired names

A. litellm-health step 7: restore .8 → gpu-dense probe (was duplicated
   to gpu-vision after 8210fd9). Add note documenting monitor key scope
   gap for gpu-dense.
B. Sweep remaining retired names presented as usable:
   - README.md:91: qwen3.6-27B-code → gpu-dense in runnable example
   - gpu-fleet.prose.md:14: qwen3.6-35B-udq4 → Carnice-Qwen3.6-MoE...
   - infrastructure-control.prose.md:226: qwen3.6-35B-udq4 → strix-moe
   - proxmox-monitor.prose.md:90: qwen3.6-35B-udq4 → strix-moe
C. Audit test: 10/10 passed (retired raw names now hard-fail)

Fix-forward from 8210fd9 (direct master push).

gpu-fleet.prose.md

@@ -11,7 +11,7 @@ description: >
   Strix Halo: strix-moe → unsloth/Qwen3.6-35B-A3B-MTP (UD-Q4_K_M, 22GB).
   RTX 5070: gpu-vision — IQ4_NL + MTP draft (~122 tok/s, 2x faster).
   UPDATED 2026-07-17: Context reduced fleet-wide from 256K to 128K for stability.
-  Strix Halo model swapped to qwen3.6-35B-udq4 (22GB, strix-moe alias).
+  Strix Halo model: Carnice-Qwen3.6-MoE-35B-A3B-Q4_K_M.gguf (alias strix-moe, 256K context).
   Instability observed near 100K at 256K (now all GPUs at 128K). 128K is the stable ceiling.
   For larger context needs → fall back to external providers (deepseek).
   VRAM headroom improved: RTX 3090 ~70%, RTX 5070 ~65%.