Add cuda wrapping flag to bisect

[mikebentley15]

After finishing #163, we will also want the wrapping functionality to be part of flit bisect. I am imagining a flag called --nvcc-wrap=<nvcc-executable> where <nvcc-executable> is either an absolute path, relative path, or executable name found in the PATH variable (as usual). This would be used for the trouble compilation. The ground-truth compilation will still follow what is in the flit-config.toml configuration file as implemented in #163 (when it is finished).

I am still unsure if we say that if you need to wrap the compilation in nvcc, then do it for all compilers. That would for sure make it easier for the original implementation. But perhaps that wouldn't be enough granularity. I'll do it on a compiler-by-compiler basis, since there may be conditional compilations and alternative compilers for compiling cuda code (clang for instance can compile some cuda kernels).