[opencv] vcpkg doesn't read CUDA_ARCH_BIN and CUDA_ARCH_PTX env vars

[nvandamme]

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

Opencv use two env vars to force specific CUDA archs and PTX version before compiling.

The actual port compile for every CUDA archs (up to 90+ gen, even if the system report a specific gen code) and thus take a very long compile time.

Using VCPKG_KEEP_ENV_VARS=CUDA_ARCH_BIN;CUDA_ARCH_PTX as a registered env var doesn't solve this issue.

Proposed solution

• read CUDA_ARCH_BIN and CUDA_ARCH_PTX env vars and set CUDA gen and ptx familly code accordingly on configure.
• modify CMake CUDA system detect to compile for the detected arch/ptx version of the current system only by default.
• OR provide a switch or an env var to force CUDA setup to match only system installed cards archs/ptx famillies

Describe alternatives you've considered

No response

Additional context

No response