[V2][RFC][wip] make room for batch eval

This moves the existing eval library to "test_suite_eval" and starts the equivalent
for batch runs. Also makes the interface a little clearer.

Essentially, the differences are:
- each metric runs on a _list_ of inputs, not just one
- each input can be paired with a reference. This is possible in the "test suite"
setup, but it is clunkier.

python/src/aiconfig/eval/api/__init__.py

@@ -9,18 +9,23 @@
     TestSuiteWithInputsSettings,
 )
 """
-from .. import common, metrics
+from .. import test_suite_common, test_suite_metrics
 
 # pyright: reportWildcardImportFromLibrary=false
-from ..lib import (
+from ..test_suite_lib import (
     TestSuiteWithInputsSettings,
     run_test_suite_outputs_only,
     run_test_suite_with_inputs,
 )
+from ..test_suite_metrics import TestSuiteMetric, brevity, substring_match
 
 __all__ = [
-    "common",
-    "metrics",
+    "TestSuiteMetric",
+    "test_suite_common",
+    "test_suite_metrics",
+    "brevity",
+    "substring_match",
+    "run_test_suite_with_inputs",
     "run_test_suite_outputs_only",
     "run_test_suite_with_inputs",
     "TestSuiteWithInputsSettings",

python/src/aiconfig/eval/batch_common.py

@@ -0,0 +1,31 @@
+from abc import abstractmethod
+from typing import Protocol, Sequence, TypeVar
+
+from aiconfig.eval import batch_common, common
+
+T_Ref = TypeVar("T_Ref")
+T_Ref_contra = TypeVar("T_Ref_contra", contravariant=True)
+
+
+class BatchEvaluationFunctionWithReference(
+    Protocol[
+        common.T_Evaluable, batch_common.T_Ref_contra, common.T_MetricValue_inv
+    ]
+):
+    @abstractmethod
+    async def __call__(
+        self,
+        data: Sequence[common.T_Evaluable],
+        ref: Sequence[batch_common.T_Ref_contra],
+    ) -> list[common.T_MetricValue_inv]:
+        pass
+
+
+class BatchEvaluationFunctionWithoutReference(
+    Protocol[common.T_Evaluable, common.T_MetricValue_inv]
+):
+    @abstractmethod
+    async def __call__(
+        self, data: Sequence[common.T_Evaluable]
+    ) -> list[common.T_MetricValue_inv]:
+        pass