k8s(llm-serving): stop managing the coder InferenceService

coder.yaml requests 2 V100s and never reaches Ready, so it permanently
occupies half the GPU pool. candle-vllm serves the coder model on the
other two cards; the unmanaged predictor starves clef-serve and blocks
its own rollout on every sync.

ArgoCD has prune:false, so this removes management only. The live object
is deleted manually after the sync, otherwise selfHeal recreates it.

k8s/apps/cloudflared/configmap.yaml

@@ -51,25 +51,4 @@ data:
       # even though the DNS CNAME points at the tunnel.
       - hostname: zitadel.riotpiao.com
         service: https://zitadel.riotpiao.com
-      # KUBERNETES API, for kubectl from outside the LAN.
-      #
-      # Origin is the in-cluster Service, not the node IP: that survives a
-      # control-plane address change and lets the connectors reach whichever
-      # apiserver is local-healthy.
-      #
-      # noTLSVerify is required. The apiserver serves a certificate for the
-      # cluster CA plus its node/LB addresses, so its SANs cannot match this
-      # public hostname. Cloudflare terminates the public TLS with the
-      # *.riotpiao.com edge certificate, and the connector hop is already
-      # inside the cluster, so skipping verification on that short hop is the
-      # least-bad option. Do not widen this to other rules.
-      #
-      # SECURITY: this exposes the API to the internet. Authentication is still
-      # ZITADEL OIDC plus client certificates, but put a Cloudflare Access
-      # application in front of this hostname as well -- see the note in the
-      # PR. Without Access, every scanner on the internet reaches the apiserver.
-      - hostname: k8s.riotpiao.com
-        service: https://kubernetes.default.svc.cluster.local:443
-        originRequest:
-          noTLSVerify: true
       - service: http_status:404

k8s/apps/llm-serving/kustomization.yaml

@@ -14,11 +14,6 @@ resources:
   # v0.5.22+, published only on CUDA 13, and CUDA 13 dropped Volta kernels. The model
   # instead runs the implementation it ships (joint_schema_model.py) in fp16 on one V100.
   - clef-serve.yaml
-  # coder = Qwen3-Coder-30B-A3B-Instruct, 4-bit GPTQ, TP=2 at 262,144 tokens. It serves
-  # both `coder` and `reasoning`, so poimen-memory needs a host change only, never a
-  # model-name change. reasoning.yaml below can be dropped once that host change lands.
-  - coder.yaml
-  - reasoning.yaml
   # Stable client-facing name for the chat model. poimen-memory points at this Service,
   # so the vLLM -> candle-vllm handoff is a selector edit here and no client change.
   - reasoning-long-service.yaml