feat: add support for objects with a model_dump method in depythonize

[pscff]

Motivation

Pydantic-style models expose a model_dump() method that returns a dict. Currently depythonize cannot deserialize these objects because they are not Mapping instances. This PR adds support for calling model_dump() automatically.

Changes

• Add model_dump_access() method to Depythonizer that:
  • Checks if the input object has a model_dump attribute
  • Calls it and expects a mapping (dict) result
  • Returns an error if the result is not a mapping
  • Returns None if the object has no model_dump method (falls through to other paths)
• Call model_dump_access() from both deserialize_map and deserialize_any, so it works for struct deserialization (serde's primary dispatch path) as well as untyped deserialization (e.g. serde_json::Value)
• Update depythonize doc comment to mention model_dump support

Test plan

• test_model_dump: deserialize a custom Python class with model_dump() returning {'x':..., 'y':...} into a Rust struct, serde_json::Value, and HashMap<String, i32>
• test_model_dump_non_mapping_error: verify that a model_dump() returning a non-mapping raises a clear error

All 89 existing + new tests pass.