fix(squad): use StringSchema instead of MultipleChoiceSchemaLower

Switch to StringSchema in the SQuAD benchmark to correctly process
extractive answers.

deepeval/benchmarks/squad/squad.py

@@ -8,7 +8,7 @@
 from deepeval.models import DeepEvalBaseLLM
 from deepeval.benchmarks.squad.task import SQuADTask
 from deepeval.benchmarks.squad.template import SQuADTemplate
-from deepeval.benchmarks.schema import MultipleChoiceSchemaLower
+from deepeval.benchmarks.schema import StringSchema
 from deepeval.telemetry import capture_benchmark_run
 from deepeval.metrics.utils import initialize_model
 
@@ -132,8 +132,8 @@ def predict(self, model: DeepEvalBaseLLM, golden: Golden) -> Dict:
 
         # Enforced model generation
         try:
-            res: MultipleChoiceSchemaLower = model.generate(
-                prompt=prompt, schema=MultipleChoiceSchemaLower
+            res: StringSchema = model.generate(
+                prompt=prompt, schema=StringSchema
             )
             prediction = res.answer
         except TypeError: