Add lint warnings for nonsensical Task configurations

tests/warning_examples/nonsensical-task-bits.yaml

@@ -0,0 +1,20 @@
+---
+- step:
+    name: train
+    image: tensorflow/tensorflow
+    command: python -u /tmp/code/my_mnist.py
+    parameters:
+      - name: A
+        type: integer
+        default: 1
+- task:
+    step: run training
+    name: task
+    type: grid-search
+    parameters:
+      - name: A
+        style: single
+        rules:
+          value: 25
+    optimization-target-metric: goodness
+    optimization-target-value: 7.2

valohai_yaml/objs/task.py

@@ -105,3 +105,13 @@ def parse(cls, data: Any) -> Task:
     def lint(self, lint_result: LintResult, context: LintContext) -> None:
         context = dict(context, task=self, object_type="task")
         lint_expression(lint_result, context, "stop-condition", self.stop_condition)
+        if self.type != TaskType.BAYESIAN_TPE:
+            for key in (
+                "optimization_target_value",
+                "optimization_target_metric",
+                "engine",
+            ):
+                if getattr(self, key) is not None:
+                    lint_result.add_warning(
+                        f"{key} only makes sense for Bayesian TPE tasks",
+                    )