Update Faithfulness rake task to use shared class

Replace the direct implementation with EvaluateAnswerFromQuestionMessage,
matching the pattern used by the Coherence and AnswerRelevancy tasks.

This also updates the spec to use the shared "a task that returns a
ScoreResult" example.

Author: chaecramb

lib/tasks/evaluation.rake

@@ -209,25 +209,15 @@ namespace :evaluation do
   task generate_faithfulness_evaluation: :environment do
     raise "Requires an INPUT env var" if ENV["INPUT"].blank?
 
-    question = Question.new(message: ENV["INPUT"], conversation: Conversation.new)
-
-    answer = AnswerComposition::PipelineRunner.call(question:, pipeline: [
-      AnswerComposition::Pipeline::SearchResultFetcher,
-      AnswerComposition::Pipeline::Claude::StructuredAnswerComposer,
-    ])
+    begin
+      result = AutoEvaluation::EvaluateAnswerFromQuestionMessage.call(
+        evaluation_class: AutoEvaluation::Faithfulness,
+        question_message: ENV["INPUT"],
+      )
 
-    if answer.status =~ /^error/
-      warn "Warning: answer has an error status: #{answer.status}"
-      abort(answer.error_message)
+      puts result.to_json
+    rescue AutoEvaluation::EvaluateAnswerFromQuestionMessage::TaskFailedError => e
+      abort e.message
     end
-
-    retrieval_context = answer.sources.used.map(&:plain_content).join("\n\n")
-
-    result = AutoEvaluation::Faithfulness.call(
-      answer_message: answer.message,
-      retrieval_context:,
-    )
-
-    puts(result.to_json)
   end
 end

spec/lib/tasks/evaluation_spec.rb

@@ -590,28 +590,9 @@
   end
 
   describe "generate_faithfulness_evaluation" do
-    it_behaves_like "an auto-evaluation generate task" do
-      let(:question_message) { "What is the current VAT rate?" }
+    it_behaves_like "a task that returns a ScoreResult" do
       let(:task_name) { "evaluation:generate_faithfulness_evaluation" }
-      let(:used_sources) do
-        [
-          build(:answer_source, used: true, chunk: build(:answer_source_chunk, plain_content: "Source 1 content")),
-          build(:answer_source, used: true, chunk: build(:answer_source_chunk, plain_content: "Source 2 content")),
-        ]
-      end
-      let(:retrieval_context) { "Source 1 content\n\nSource 2 content" }
-
-      before do
-        allow(answer.sources).to receive(:used).and_return(used_sources)
-
-        allow(AutoEvaluation::Faithfulness)
-          .to receive(:call)
-          .with(
-            answer_message: answer.message,
-            retrieval_context:,
-          )
-          .and_return(evaluation_result)
-      end
+      let(:evaluation_class) { AutoEvaluation::Faithfulness }
     end
   end
 end