[PATCH roland/lll-futex] Separate Linuxisms from lowlevellock.h, make a generic one

[Bernie Ogden]

> Perhaps there should be a test that we get EOWNERDEAD if a lock holder dies in an unclean way before the call to __pthread_mutex_trylock - I didn't see one and I guess that would be doable with a pair of threads. But that's getting further away from issues relating to this patch.

Actually there is already a test for EOWNERDEAD (tst-robust2 and
tst-robust5). Just not in the racy case.