[CLIENTS]fix cuda_runtime not found and assert len(MARKERS) >= len(y_keys[index])

[engineer1109]

When compile with

cmake .. -DCLIENTS=ON -DCUBLAS=ON
make clblast_client_xgemm

And Run

python3 ../scripts/benchmark/benchmark.py  -p 0 -d 0 -b gemm -c clBLAS CPU-BLAS cuBLAS

Will meet 2 bugs:

1. Header cuda_runtime.h not found
   add ${CUBLAS_INCLUDE_DIRS} to REF_INCLUDES to fix this.

2. assert len(MARKERS) >= len(y_keys[index])

python3 ../scripts/benchmark/benchmark.py  -p 0 -d 0 -b gemm -c clBLAS CPU-BLAS cuBLAS

The cuBLAS lacks the MARKERS.
Changed:
MARKERS = ["o-", "x-", ".-", "--"]
to give cuBLAS the markers.

[engineer1109]

@CNugteren BUGS with CLIENTS benchmark small fix.