[ENH] add list_metrics and compute_metric MCP tools (#79)

[rupeshca007]

Summary

Closes #79
Adds two new MCP tools that expose sktime's forecasting performance metrics to LLM agents:

• list_metrics - returns a catalogue of all supported metrics (name, description, lower_is_better, scale_dependent, requires_y_train).
• • compute_metric - evaluates any metric from the catalogue on explicit y_true / y_pred list inputs. Scale-normalised metrics (MASE, RMSSE) additionally accept a y_train argument.

Supported Metrics

Name	Description	Scale-dependent	Requires y_train
MAE	Mean Absolute Error	yes
MSE	Mean Squared Error	yes
RMSE	Root Mean Squared Error	yes
MAPE	Mean Absolute Percentage Error
SMAPE	Symmetric MAPE
MASE	Mean Absolute Scaled Error		yes
RMSSE	Root Mean Squared Scaled Error		yes
MedAE	Median Absolute Error	yes
MedSE	Median Squared Error	yes
MedRMSE	Median Root Mean Squared Error	yes
MedAPE	Median Absolute Percentage Error
MedSAPE	Symmetric Median APE
GMAE	Geometric Mean Absolute Error	yes
GMSE	Geometric Mean Squared Error	yes
GRMSE	Geometric Root Mean Squared Error	yes

Files Changed

File	Purpose
src/sktime_mcp/tools/metrics.py	New file - tool implementations
src/sktime_mcp/tools/__init__.py	Export new tools
src/sktime_mcp/server.py	Register tool schemas + dispatch
tests/test_metrics.py	22 tests (8 pass locally, 14 skipped when sktime not installed; all 22 run on CI)

How LLMs Can Use This

# Step 1: discover what metrics exist
list_metrics()

# Step 2: evaluate a forecast
compute_metric(metric="RMSE", y_true=[100,110,120], y_pred=[98,112,118])
# -> {"success": true, "metric": "RMSE", "value": 2.16, "lower_is_better": true}

# Step 3: MASE needs training data for normalisation
compute_metric(metric="MASE", y_true=[...], y_pred=[...], y_train=[...])

Checklist

• New tool file follows existing patterns (tools/describe_estimator.py, tools/evaluate.py)
• - [x] Registered in list_tools() with full JSON schema (including required fields)
• - [x] Dispatch added to call_tool()
• - [x] Tests written for both happy-path and error cases
• - [x] pytest.mark.skipif used (not importorskip) so list_metrics tests always run without sktime installed
• [ ]