[WIP/Review] Add basic CUDA support (for CMake projects)

[rietmann-nv]

The way CMake compiles cuda files happens in a way that makes it hard for ccls to parse the command and extract the relevant flags to make clang work correctly. I've added a few changes to ccls to add explicit CUDA support in both the file detection and .ccls configuration to make basic things like code completion and jump-to-definition mostly work.

The nvcc command looks like this (notice two step command)

/usr/local/cuda/bin/nvcc -ccbin=gcc-6  -I../src -I../external/cutlass -I../external/cub -isystem=../external/googletest/googletest/include   -Xcompiler -fopenmp --expt-extended-lambda --std=c++11 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_70,code=compute_70 -g   -x cu -c /home/max/dev/cuml/ml-prims/test/add.cu -o test/CMakeFiles/mlcommon_test.dir/add.cu.o && /usr/local/cuda/bin/nvcc -ccbin=gcc-6  -I../src -I../external/cutlass -I../external/cub -isystem=../external/googletest/googletest/include   -Xcompiler -fopenmp --expt-extended-lambda --std=c++11 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_70,code=compute_70 -g   -x cu -M /home/max/dev/cuml/ml-prims/test/add.cu -MT test/CMakeFiles/mlcommon_test.dir/add.cu.o -o $DEP_FILE

A similar clang command should look something like this:

clang -I../src -I../external/cutlass -I../external/cub -isystem=../external/googletest/googletest/include --std=c++11 --cuda-gpu-arch=sm_70 --cuda-path=/usr/local/cuda-9.2/ -c add.cu

I wanted this for myself and my work, but I know there are others who would like this as well. I wanted to make sure to contribute back, given how much I've been using ccls in the last year. Most of the additions are pretty simple, but the changes I made at project.cc:437 feel a bit hacky, and I wanted your feedback to streamline them into something better, if that is possible. Basically I implemented a whitelist which only allows (-I, -isystem, -std=) type flags. This basically works, but maybe you have a better idea.

Additionally it requires the user to make their .ccls look something like:

%compile_commands.json
%cu --cuda-gpu-arch=sm_70
%cu --cuda-path=/usr/local/cuda-9.2/

The --cuda-gpu-arch=sm_70 and --cuda-path=/usr/local/cuda-9.2/ could possibly be added automatically.

Anyway, thanks for ccls and I look forward to your feedback.