fix: raise error for missing explicit exogenous columns

[Saloni-0465]

Reference Issues/PRs

N/A

What does this implement/fix? Explain your changes.

This PR fixes silent dropping of explicitly requested exogenous columns during data conversion.

Previously, if a config specified exog_columns=["promo", "holiday"] but only promo existed in the loaded data, to_sktime_format would silently use the available column and ignore the missing one. This could mislead users or LLM agents into thinking all requested covariates were used.

This PR updates DataSourceAdapter.to_sktime_format to validate explicit exog_columns before constructing X. If any requested exogenous column is missing, it raises a clear ValueError listing the missing columns and available columns. When all requested columns exist, their order is preserved exactly.

Does your contribution introduce a new dependency? If yes, which one?

No.

What should a reviewer concentrate their feedback on?

• Whether missing explicit exog_columns should fail fast rather than warn
• Clarity of the error message returned to users and agents
• Whether the validation belongs in DataSourceAdapter.to_sktime_format

Any other comments?

This is a small data-contract safety fix for custom data workflows. It helps prevent agents from making forecasting decisions under the false assumption that missing exogenous variables were used.

Validation:

• python -m compileall src/sktime_mcp/data/base.py tests/test_data_exog_column_validation.py
• Manual check: missing holiday in exog_columns raises a clear ValueError
• Manual check: Executor.load_data_source returns success=False with error_type="ValueError"
• Manual check: valid explicit exogenous columns are preserved in order

PR checklist

For all contributions

• I've added myself to the list of contributors.
• Optionally, I've updated CODEOWNERS.
• I've added unit tests and made sure they pass locally where possible.

For new estimators

• Not applicable