Investigate NVCC flag '-forward-unknown-to-host-compiler' on by default

https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#options-for-guiding-compiler-driver-forward-host-compiler

This seems extremely useful to always turn on.

Start with the linked NVIDIA NVCC documentation section on forwarding options, then inspect BLT's build configuration to find where NVCC flags are defined. Determine whether enabling '-forward-unknown-to-host-compiler' by default is compatible with supported toolchains and identify the validation needed; done means a justified default decision and corresponding build coverage.

Written by the indexing model from the issue text.