Fix OTel collector advertise host in direct Kubernetes mode

[mconflitti-pbc]

Summary

• Removes the default that set OpenTelemetry.CollectorAdvertiseHost to the Connect Service DNS name in the gcfg config — the ephemeral OTel collector port is not exposed on the Service, causing content job pods to time out on every log send
• Injects status.podIP via the Kubernetes Downward API as CONNECT_OPENTELEMETRY_COLLECTORADVERTISEHOST when config.OpenTelemetry.Enabled: true — the OTel collector already binds to 0.0.0.0 in off-host mode so direct pod-to-pod routing works
• Adds tests/otel_test.yaml to cover the new behavior

Validation

Env var is set is on connect pod

➜  otelcol-advertise-test git:(otelcol-advertise-test) ✗ PULUMI_ARCH=aws/k8s PULUMI_STACK=otelcol-advertise-test just kubectl exec -it deployment/rstudio-connect-prod -- env | grep CONNECT_OPENTELEMETRY
uv run -- pulumi -C architectures/aws/k8s -s otelcol-advertise-test stack output kubeconfig
CONNECT_OPENTELEMETRY_COLLECTORADVERTISEHOST=10.11.128.216

Proper host ip is passed into content pod:

Environment:
      USER:                            rstudio-connect
      USERNAME:                        rstudio-connect
      LOGNAME:                         rstudio-connect
      HOME:                            /tmp
      TMPDIR:                          /tmp
      PICOTEL_PREFIX:                  PICOTEL
      PICOTEL_SERVICE_NAME:            posit-connect-python
      PICOTEL_EXPORTER_OTLP_ENDPOINT:  http://10.11.128.216:32845/ <<<<<<<<<
      PICOTEL_TRACEPARENT:             00-4272be7a485e5338f7f21f268ed59265-2a45a203df4d0f60-01
      PICOTEL_RESOURCE_ATTRIBUTES:     content.guid=ac5d7e36-b2ed-4d80-ad8b-1bfa155f5182,content.id=7,content.bundle.id=8,job.key=ckTIUYqrLnGdCrwr

No picotel errors in content logs

Test plan

• Deploy Connect on Kubernetes with OpenTelemetry.Enabled: true
• Verify CONNECT_OPENTELEMETRY_COLLECTORADVERTISEHOST is set to the pod IP in the Connect pod
• Deploy content requiring a fresh Python environment restore and confirm the job completes without stalling

[mconflitti-pbc]

This may end up just becoming a reference as a workaround until we fix Connect to get the ip on its own:

➜  otelcol-advertise-test git:(otelcol-advertise-test) ✗ PULUMI_ARCH=aws/k8s PULUMI_STACK=otelcol-advertise-test just kubectl exec -it deployment/rstudio-connect-prod -- env | grep CONNECT_OPENTELEMETRY
uv run -- pulumi -C architectures/aws/k8s -s otelcol-advertise-test stack output kubeconfig
CONNECT_OPENTELEMETRY_COLLECTORADVERTISEHOST=10.11.128.216

[dbkegley]

Looks right to me. Just to confirm what I'm reading - if we set config.OpenTelemetry.CollectorAdvertiseHost explicitly in the config then the env var is not added to the deployment, right?