kernel: Force disable LLVM CUDA

The kernel tries to define the LINUX_COMPILER macro from the output
of `$(CC) -v | grep ' version '`, but in clang that might produce
more than one line containing ' version ' when CUDA is installed:

```
Android (8490178, based on r450784d) clang version 14.0.6
Found CUDA installation: /usr/local/cuda, version 11.0
```

That causes the macro to expand into multiple lines, upsetting the
compiler that cannot find the ending double quote.

In upstream linux this is fixed from v5.8

Change-Id: Icde4253e989df1cbbefc0ce98f437d16822fc93b

build/tasks/kernel.mk

@@ -239,7 +239,7 @@ ifneq ($(TARGET_KERNEL_CLANG_COMPILE),false)
     endif
     PATH_OVERRIDE += PATH=$(TARGET_KERNEL_CLANG_PATH)/bin:$$PATH
     ifeq ($(KERNEL_CC),)
-        KERNEL_CC := CC="$(CCACHE_BIN) clang"
+        KERNEL_CC := CC="$(CCACHE_BIN) clang --cuda-path=/dev/null"
     endif
 endif