fix(metadata): Patching when receiving entry is not a dict (#338)

## Description
Fixing an edge case in the `external_graph` runner when used with a LAM
model. When using this runner with the `graph_dataset` option, it will
patch the `dataloader.dataset` entry:


https://github.com/ecmwf/anemoi-inference/blob/196aa55a57204c4bcb823e7b6546d79505f5eaee/src/anemoi/inference/runners/external_graph.py#L189-L199

But if the existing `dataloader.dataset` entry in the metadata happens
to be a list of datasets, the patching crashes. When patching an
existing entry, the patch function did not check if the receiving entry
is also a dictionary. In the case of a list on the receiving side, it
would try to index the list with the incoming dictionary keys. This PR
fixes that.

Also adds a test for this function.


***As a contributor to the Anemoi framework, please ensure that your
changes include unit tests, updates to any affected dependencies and
documentation, and have been tested in a parallel setting (i.e., with
multiple GPUs). As a reviewer, you are also responsible for verifying
these aspects and requesting changes if they are not adequately
addressed. For guidelines about those please refer to
https://anemoi.readthedocs.io/en/latest/***

By opening this pull request, I affirm that all authors agree to the
[Contributor License
Agreement.](https://github.com/ecmwf/codex/blob/main/Legal/contributor_license_agreement.md)

Author: gmertes

src/anemoi/inference/metadata.py

@@ -1129,7 +1129,7 @@ def patch(self, patch: dict) -> None:
         def merge(main: dict[str, Any], patch: dict[str, Any]) -> None:
 
             for k, v in patch.items():
-                if isinstance(v, dict):
+                if isinstance(v, dict) and isinstance(main.get(k, {}), dict):
                     if k not in main:
                         main[k] = {}
                     merge(main[k], v)

tests/unit/test_metadata.py

@@ -0,0 +1,54 @@
+# (C) Copyright 2025 Anemoi contributors.
+#
+# This software is licensed under the terms of the Apache Licence Version 2.0
+# which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
+#
+# In applying this licence, ECMWF does not waive the privileges and immunities
+# granted to it by virtue of its status as an intergovernmental organisation
+# nor does it submit to any jurisdiction.
+
+import pytest
+
+from anemoi.inference.metadata import Metadata
+from anemoi.inference.testing.mock_checkpoint import mock_load_metadata
+
+
+@pytest.mark.parametrize(
+    "initial, patch, expected",
+    [
+        (
+            {"config": {"dataloader": {"dataset": "abc"}}},
+            {"config": {"dataloader": {"something": {"else": "123"}}}},
+            {"dataset": "abc", "something": {"else": "123"}},
+        ),
+        (
+            {"config": {"dataloader": [{"dataset": "abc"}, {"dataset": "xyz"}]}},
+            {"config": {"dataloader": {"cutout": [{"dataset": "123"}, {"dataset": "456"}]}}},
+            {"cutout": [{"dataset": "123"}, {"dataset": "456"}]},
+        ),
+        (
+            {"config": {"dataloader": "abc"}},
+            {"config": {"dataloader": "xyz"}},
+            "xyz",
+        ),
+    ],
+)
+def test_patch(initial, patch, expected):
+    metadata = Metadata(initial)
+    assert metadata._metadata == initial
+
+    metadata.patch(patch)
+    assert metadata._metadata["config"]["dataloader"] == expected
+
+
+def test_constant_fields_patch():
+    model_metadata = mock_load_metadata("unit/checkpoints/atmos.json", supporting_arrays=False)
+    metadata = Metadata(model_metadata)
+
+    fields = ["z", "sdor", "slor", "lsm"]
+    metadata.patch({"dataset": {"constant_fields": fields}})
+    assert metadata._metadata["dataset"]["constant_fields"] == fields
+
+    # check that the rest of the metadata is still the same after patching
+    metadata._metadata["dataset"].pop("constant_fields")
+    assert model_metadata == metadata._metadata