Add majority_k helper + tests

[Ardbiu]

This PR introduces majority_k — a lightweight helper that wraps an existing DSPy predictor, runs it k times with identical inputs, and returns the majority result using the existing majority() function.

This helper reduces repetitive boilerplate when sampling multiple completions via the same predictor.

Details
• Supports both callable predictors and existing prediction/completion objects
• Does not modify core prediction logic
• For callable predictors, wraps and returns a callable
• For non-callable inputs, directly forwards to majority()

Tests
Added tests in tests/predict/test_majority_k.py:

1. callable predictor → called k times
2. passing existing completions → forwarded directly
3. kwargs forwarded correctly

Implements feature requested in Issue #8988