Initial TRT bindings

[wizenink]

This PR aims to add support for TensorRT to cudarc.

As TensorRT is a C++ only lib, bindgen automatic generation does not work here, so a cpp wrapper is created to flatten TensorRT into C code.

Right now, we are not downloading redist files from nvidia, that would be a big TODO.

Working ATM:

• Some basic tests
• TensorRT linked and working with CUDA 13 and CUDA 12

Implementation details:

• TensorRT thin wrapper always statically linked (Very basic wrapper, I dont see the benefit of trying to dynamic load it, and CC only supports static lib generation)
• TensorRT code not being fetched on bindings_generator

Missing:

• Feature gating TRT code in the cpp wrapper depending on the cuda version, maybe passing some defines onto the header coming from rust feature flags
• Expand the trt wrapper to expose more TRT features, this is only a small POC
• Fix some deprecation warnings on the TRT plugin code
• Safe wrappers (This is a big one)

@chelsea0x3b maybe you can check it out and just check if the implementation logic is right or if maybe another path should be taken to implement this

[chelsea0x3b]

@wizenink honestly this looks much cleaner than I expected, I think we could totally move forward with this

[wizenink]

Nice :) Will focus on having feature gates synced between rust and the C wrapper for now, and leave fetching trt redist automatically for the future