fetch_supported_metrics() fails at import due to inadequate dummy data

[tanmaygarg2911]

Problem: The function tries to detect supported sklearn metrics by calling them with dummy data, but:

• Only catches TypeError (missing: ValueError, AttributeError, AxisError)
• Uses Python lists instead of numpy arrays
• Uses 1D arrays, but classification metrics need 2D probability arrays
• Uses integer dtype, but some metrics require float

Impact: Library cannot be imported at all.

Root cause: Metrics have diverse input requirements that a single dummy dataset cannot satisfy.

Proposed solutions:

1. Use multiple dummy datasets (1D regression, 2D classification probabilities)
2. Catch broader exception types

I can submit a PR with the approach you prefer.