[Feature] Add built-in sktime tool for time series forecasting workflows

[Saloni-0465]

Summary

I'd like to add a built-in sktime tool to ml-intern that lets the agent
run time series ML workflows directly, without requiring an external MCP server.

Motivation

sktime is a widely-used Python framework for time series tasks (forecasting,
classification, anomaly detection). Right now, if a user asks ml-intern to
"forecast this time series using ARIMA" or "compare forecasting models on
airline data", the agent has no direct way to do this.

A built-in sktime tool would let the agent:

• Discover available estimators by task type
• Get detailed parameter info for any estimator
• Run end-to-end forecasting on built-in datasets
• Evaluate and compare model accuracy

Proposed Design

Following the same pattern as papers_tool.py, I'd add:

• agent/tools/sktime_tool.py — tool spec + handler with operations:
  • list_estimators — browse estimators filtered by task/tags
  • describe_estimator — get parameters and capabilities
  • forecast — fit and predict on built-in datasets
  • list_datasets — show available datasets

Why I'm the right person for this

I'm an active contributor to sktime-mcp (9 PRs) and have deep familiarity
with sktime's registry and API. I have a working implementation ready.

Happy to open a PR if this direction looks good to maintainers!

[fglogan]

closed per maintainer request

[Saloni-0465]

Thanks for the response! @fglogan
Could you share a bit of guidance on the preferred approach? I'm happy to redirect my contribution.