Honor cpu:null on the agent container resources (#334)

A consumer-supplied cpu:null on agents[].resources.limits was rendering
into the AmazonCloudWatchAgent CR as a literal cpu: null, which is invalid
under the CRD, instead of removing the cpu limit. The agent resources go
through mergeOverwrite over the chart defaults, and sprig's mergeOverwrite
keeps nil source values, so the null survived into the merged map.

Add a cloudwatch-agent.pruneNulls helper that strips nil leaves from the
merged resources map before rendering, so cpu:null removes the limit
(matching how the fluent-bit DaemonSet, which never merges over a default,
already behaves).

Author: movence

charts/amazon-cloudwatch-observability/templates/_helpers.tpl

@@ -784,3 +784,20 @@ winning on key collision.
 {{- end -}}
 {{- $merged | toYaml -}}
 {{- end -}}
+
+{{/* Recursively drop nil leaves so a user-supplied `cpu: null` removes the limit instead of emitting literal null. mergeOverwrite keeps nil values from the default, so prune after merge. */}}
+{{- define "cloudwatch-agent.pruneNulls" -}}
+{{- $in := . -}}
+{{- $out := dict -}}
+{{- range $k, $v := $in -}}
+  {{- if kindIs "map" $v -}}
+    {{- $nested := include "cloudwatch-agent.pruneNulls" $v | fromYaml -}}
+    {{- if $nested -}}
+      {{- $_ := set $out $k $nested -}}
+    {{- end -}}
+  {{- else if not (kindIs "invalid" $v) -}}
+    {{- $_ := set $out $k $v -}}
+  {{- end -}}
+{{- end -}}
+{{- $out | toYaml -}}
+{{- end -}}

charts/amazon-cloudwatch-observability/templates/linux/cloudwatch-agent-custom-resource.yaml

@@ -129,7 +129,8 @@ spec:
       enabled: {{ $agent.prometheus.targetAllocator.prometheusCR.enabled | default false }}
     {{- end }}
   {{- end }}
-  {{- with $agent.resources }}
+  {{- $prunedResources := include "cloudwatch-agent.pruneNulls" $agent.resources | fromYaml }}
+  {{- with $prunedResources }}
   resources: {{- toYaml . | nindent 4}}
   {{- end }}
   volumeMounts: