[ENH] expose explicit data shape metadata for agent reasoning #416

Summary

load_data_source and load_data_source_async currently return basic metadata such as rows, columns, dates, and dtypes, but they do not explicitly tell an agent whether a loaded handle is univariate or multivariate, whether exogenous variables are present, or what kind of index backs the series.

Why this matters

Agentic workflows often need to decide which estimator family or tool path to take before fitting. Right now an agent has to infer too much from raw column names and dtypes.

Concrete gaps:

• univariate vs multivariate target is not explicit
• exogenous presence is not explicit
• exogenous width is not explicit
• index type is not explicit (datetime vs period vs range / integer)

This makes tool selection and model routing harder than it needs to be.

Proposal

Augment data-handle metadata with explicit agent-friendly fields, for example:

• target_scitype
• target_variates
• has_exog
• exog_variates
• index_type
• n_target_columns
• n_exog_columns

Expected benefit

This would make loaded data self-describing for LLM / MCP agents and reduce avoidable reasoning errors when selecting estimators or deciding whether exogenous-aware workflows are available.