[FEA]: Add SM90 to architecture list for CI

### Is this a duplicate?

- [X] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cccl/issues) for this request and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md)

### Area

Infrastructure

### Is your feature request related to a problem? Please describe.

As a user and maintainer of CCCL, I want to be sure that CCCL works when compiled and run for SM90 (Hopper). However, CCCL's CI does not currently compile (or run) for SM90.

There are a number of things that complicate simply adding SM90 to the architecture list:

- Compiling RDC tests in Thrust/CUB break for SM90 today: https://github.com/NVIDIA/cccl/issues/219
- The CUDA architecture list is [hard-coded in the CMake preset](https://github.com/NVIDIA/cccl/blob/main/CMakePresets.json#L16) and there's no longer a way to customize the GPU architectures in the [CI matrix](https://github.com/NVIDIA/cccl/blob/7395463758994ff95f2d583fb7d537fbbbf1c262/ci/matrix.yaml#L28-L51). This functionality used to exist, but was removed with the addition the CMake preset.
- Simply adding SM90 to the build list will still not exercise compiling (or running) SM90 specific tests in libcudacxx due to the [current design of libcudacxx's testing infrastructure](https://github.com/NVIDIA/cccl/issues/1084)

### Describe the solution you'd like

### Tasks
- [x] Enable specifying CUDA architecture per CI job
- [ ] https://github.com/NVIDIA/cccl/issues/1137
- [ ] https://github.com/NVIDIA/cccl/issues/219
- [ ] https://github.com/NVIDIA/cccl/issues/1084

### Describe alternatives you've considered

_No response_

### Additional context

_No response_