[Feature] Add float32 zero-copy angle encoding in qdp-core (single sample)

[viiccwen]

Reference: #1003

Today, Mahout only supports the float32 zero-copy path for amplitude encoding. Angle encoding still requires float64 on the GPU-pointer path.

This issue covers these parts:

• Rust core API support for single-sample angle f32
• CUDA kernel support for single-sample angle f32
• Core-level tests for success/error paths
• Kernel-level numerical validation