[Inference API] Fix node local rate limit calculator tests for non-snapshot builds (#121527)

Author: timgrein

muted-tests.yml

@@ -370,8 +370,6 @@ tests:
 - class: org.elasticsearch.backwards.MixedClusterClientYamlTestSuiteIT
   method: test {p0=search.vectors/42_knn_search_int4_flat/Vector similarity with filter only}
   issue: https://github.com/elastic/elasticsearch/issues/121412
-- class: org.elasticsearch.xpack.inference.common.InferenceServiceNodeLocalRateLimitCalculatorTests
-  issue: https://github.com/elastic/elasticsearch/issues/121294
 - class: org.elasticsearch.xpack.ml.integration.ClassificationIT
   method: testDependentVariableIsAliasToKeyword
   issue: https://github.com/elastic/elasticsearch/issues/121492

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/common/InferenceServiceNodeLocalRateLimitCalculatorTests.java

@@ -33,6 +33,11 @@ public class InferenceServiceNodeLocalRateLimitCalculatorTests extends ESIntegTe
 
     public void setUp() throws Exception {
         super.setUp();
+        assumeTrue(
+            "If inference_cluster_aware_rate_limiting_feature_flag_enabled=false we'll fallback to "
+                + "NoopNodeLocalRateLimitCalculator, which shouldn't be tested by this class.",
+            InferenceAPIClusterAwareRateLimitingFeature.INFERENCE_API_CLUSTER_AWARE_RATE_LIMITING_FEATURE_FLAG.isEnabled()
+        );
     }
 
     public void testInitialClusterGrouping_Correct() throws Exception {