There is no vDSO support with Valgrind. Detect when running on top of
such virtual machine using the Valgrind API, picking the fallback
clock_gettime() implementation if so.

This is not a problem since the EVL core does intercept and honor this
syscall directly from the out-of-band stage as needed. The incurred
overhead is negligible compared to what Valgrind already adds anyway.

Also, provide a placeholder for valgrind/valgrind.h which should be
picked when Valgrind is not installed on the build system.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>