Core-LB-G05: Document least-loaded semantics

[RicheyWorks]

Overall Classification: WARN / core-feature-audit

Campaign: Core LoadBalancer Reliability Contract Campaign
Goal: Core-LB-G05 - Least-loaded semantics decision
Scope: docs/test-only contract characterization

Summary

• Adds focused least-loaded facade contract tests.
• Preserves the current positive-load behavior: leastLoaded sorts healthy servers by load score internally, but allocates equal shares across the healthy server set for the tested positive-load cases.
• Characterizes current zero-load behavior: leastLoaded exposes the lowest-load candidate before the current loop stops.
• Compares leastLoaded positive-load shape with roundRobin where useful.
• Records the G05 decision in the core load-balancer feature contract.
• Does not start Core-LB-G06.

Changed Files

• docs/agent/CORE_LOADBALANCER_FEATURE_CONTRACT.md
• src/test/java/com/richmond423/loadbalancerpro/core/CoreLoadBalancerLeastLoadedSemanticsTest.java

Verification

• mvn test "-Dtest=CoreLoadBalancerLeastLoadedSemanticsTest" passed: 5 tests, 0 failures, 0 errors, 0 skipped.
• mvn test "-Dtest=AgentCoreLoadBalancerFeatureContractDocumentationTest" passed: 7 tests, 0 failures, 0 errors, 0 skipped.
• mvn test "-Dtest=CoreLoadBalancerLeastLoadedSemanticsTest,CoreLoadBalancerCrossStrategyEdgeInvariantTest,CoreLoadBalancerCapacityPredictiveOverloadTest,CoreLoadBalancerDeterministicOrderingTieTest,LoadBalancerTest,AgentCoreLoadBalancerFeatureContractDocumentationTest" passed: 98 tests, 0 failures, 0 errors, 0 skipped.
• mvn -q test passed.
• mvn -q "-DskipTests" package passed.
• mvn -B package passed: 2,506 tests, 0 failures, 0 errors, 0 skipped.
• git diff --check passed.
• git diff --cached --check passed.
• git diff --check origin/main...HEAD passed.
• .\scripts\smoke\enterprise-lab-workflow.ps1 -Package passed.

Safety

• No production code changed.
• No Maven config changed.
• No CI/workflow changed.
• No Dockerfile/Compose changes.
• No scripts/runtime resources/endpoints/secrets/external targets/deployment/automation changed.
• No tests weakened.
• PR Add next-level architecture skeleton #334 was not touched and was not used as a base.

Not-Proven Boundaries

• No production readiness/certification.
• No live-cloud or real-tenant validation.
• No runtime enforcement claim beyond implemented tests.
• No load/stress/benchmark evidence.
• No throughput/p95/p99 production evidence.
• No real-world latency improvement claim.
• No replay/evidence/report/storage/export proof.
• No broader automation.