Remove `cudaDeviceSynchronize`

• CUDA's default synchronization w/ the CPU is enough s.t. we don't need cudaDeviceSynchronize.

hercules_cg/src/gpu.rs

@@ -661,12 +661,6 @@ extern \"C\" {} {}(",
             w,
             "\tif (cudaSuccess != err) {{ printf(\"Error1: %s\\n\", cudaGetErrorString(err)); }}\n"
         )?;
-        write!(w, "\tcudaDeviceSynchronize();\n")?;
-        write!(w, "\terr = cudaGetLastError();\n")?;
-        write!(
-            w,
-            "\tif (cudaSuccess != err) {{ printf(\"Error2: %s\\n\", cudaGetErrorString(err)); }}\n"
-        )?;
         if has_ret_var {
             // Copy return from device to host, whether it's primitive value or collection pointer
             write!(w, "\t{} host_ret;\n", ret_type)?;