Update agent_next_test.go and DESIGN.md for improved test reliability and documentation clarity

- Modified the test case in agent_next_test.go to use a deterministic sub-scenario, reducing flakiness in tests caused by unpredictable LLM responses.
- Enhanced DESIGN.md to clarify the structure and organization of the search module, including detailed descriptions of new types, interfaces, and configuration options.
- Updated the documentation to reflect changes in the citation system, source weighting, and the overall architecture of the search module, ensuring consistency and better understanding for developers.

agent/assistant/agent_next_test.go

@@ -168,7 +168,9 @@ func TestAgentNextConditional(t *testing.T) {
 
 	ctx := newAgentNextTestContext("test-conditional", "tests.realworld-next")
 	messages := []context.Message{
-		{Role: context.RoleUser, Content: "scenario: conditional - Task completed"},
+		// Use conditional_success sub-scenario for deterministic behavior
+		// This avoids test flakiness caused by LLM response unpredictability
+		{Role: context.RoleUser, Content: "scenario: conditional_success - Task completed"},
 	}
 
 	response, err := agent.Stream(ctx, messages)