bug: model_variable lost during CMORizer rule expansion with DefaultPipeline

[pgierz]

Problem

When using DefaultPipeline with a config that sets model_variable: "temp" and cmor_variable: "thetao", the get_variable pipeline step fails with:

KeyError: "No variable named 'thetao'. Variables on the dataset include ['temp', ...]"

The model_variable attribute is correctly set during Rule.from_dict() (verified locally), but by the time get_variable runs, rule_spec.model_variable returns "thetao" (the cmor_variable) instead of "temp".

Reproduction

Happens in CI with the pycmor-test-data-fesom external package's config_cmip6_fesom_2p6.yaml:

rules:
  - name: "thetao_with_levels"
    cmor_variable: "thetao"
    model_variable: "temp"
    pipelines:
      - default
pipelines:
  - name: default
    uses: pycmor.core.pipeline.DefaultPipeline

Hypothesis

The model_variable dynamic attribute is likely lost during one of:

• _rules_expand_drvs() -- creates new Rule objects via deep copy
• _rules_depluralize_drvs() -- may reconstruct rules
• Rule pickling/unpickling through Prefect task submission

Context

Discovered during #253 integration testing. All fesom_2p6 tests with DefaultPipeline fail. This is pre-existing behavior -- the external package is the first to exercise DefaultPipeline end-to-end with model_variable != cmor_variable.

Related

• PR feat: test infrastructure + tutorial dataset system #253 (test infrastructure + tutorial system)
• External package: https://github.com/FESOM/pycmor_test_data