[Feature][QDP] Add pluggable backend dispatch for Python-exported kernels

[400Ping]

Background

We want a single API surface for kernels exported from Rust to Python, while allowing multiple backend implementations underneath.

Proposal

Define one public API and add a backend parameter to dispatch to different implementations, for example:

• triton
• cuda
• future backends if needed

This keeps the Python integration stable while letting us compare implementations and evolve performance over time.

Goals

• Export a consistent kernel API from Rust to Python
• Support multiple backend implementations behind the same interface
• Use Triton as the initial baseline implementation
• Allow CUDA as a higher-performance path where available
• Make backend selection explicit and extensible

[yehfelareborn]

I would like to contribute to this issue, thanks

[400Ping]

I would like to contribute to this issue, thanks

This should be implemented when I finished up supporting AMD GPUs

[yehfelareborn]

ok, thanks

[400Ping]

Closing this since this is implemented in #1158