Detect nested Keras Lambda layers

[dfgvaetyj3456356-hash]

Summary

• Recursively traverse Keras model config trees when looking for Lambda layers.
• Reuse the same traversal for .keras and H5 Keras configs.
• Add dependency-light regression coverage for nested Lambda detection, including an in-memory .keras archive scanner path.

Why

The current Keras Lambda checks only inspect the first-level config.layers array. Keras model configs can contain nested model/layer configs, so a nested Lambda layer can be missed by ModelScan even though top-level Lambda layers are reported.

This is related to #340's nested config work, but narrower: #340 focuses on unsafe module references and preserves the existing top-level Lambda check. This PR specifically makes Lambda detection recursive and applies the same helper to H5 config parsing.

Testing

• python -m pytest tests\test_keras_nested_lambda_scan.py -q
• python -m compileall modelscan\scanners\keras_utils.py modelscan\scanners\keras\scan.py modelscan\scanners\h5\scan.py tests\test_keras_nested_lambda_scan.py

I could not run the broader ests/test_modelscan.py targets locally because the checkout is missing optional test dependencies such as dill and TensorFlow.