fix: correct metric names in llama-swap monitoring (llamacpp_* -> llamaswap_*), update alerts + dashboard + scrape config

cluster/applications/monitoring/values.yaml

@@ -299,7 +299,10 @@ prometheus:
 
       # llama-swap (production, since 2026-08-18)
       # Replaces the per-model /metrics?model=<id> jobs above (all targeting now-deprecated :8002).
-      # llama-swap natively exposes /metrics on its own endpoint with model-labeled metrics.
+      # llama-swap exposes system-level metrics (llamaswap_*) — VRAM, GPU util, power, CPU, network.
+      # Per-model inference metrics (tokens/sec, latency, KV-cache) are NOT exposed at the proxy level;
+      # they remain on the individual llama-server child instances, scraped via node_exporter textfile
+      # collector for VRAM, and via the GPU exporter (:9835) for GPU-level telemetry.
       - job_name: llama-swap
         scrape_interval: 30s
         scrape_timeout: 10s
@@ -314,7 +317,7 @@ prometheus:
         honor_labels: true
         metric_relabel_configs:
           - source_labels: [__name__]
-            regex: 'llamacpp_.*'
+            regex: 'llamaswap_.*'
             action: keep
 
 # ─── Grafana ──────────────────────────────────────────────────────────────────