feat: Add float32 zero-copy angle encoding in Python bindings (single sample)

[viiccwen]

Related Issues

Closes #1250

Changes

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Why

QDP core already supports single-sample angle encoding from CUDA float32
buffers, but the Python binding did not expose that path. This PR adds the
Python-side support while keeping batch CUDA angle inputs on float64.

How

• route 1D CUDA float32 angle tensors through the dedicated zero-copy angle
  GPU-pointer path
• keep CUDA angle float32 support limited to 1D single-sample tensors
• add targeted Python tests for success and rejection paths
• document the CUDA tensor dtype rules for amplitude and angle
• pin the IQP naive-reference comparison test to float64 precision so the
  comparison matches the reference dtype

Checklist

• Added or updated unit tests for all changes
• Added or updated documentation for all changes