shmem_wait_until_all[_vector] implementation looks wrong

[dalcinl]

I cannot make sense of the implementation of shmem_wait_until_all[_vector]().
https://github.com/openshmem-org/osss-ucx/blob/main/src/shmemc/ucx/waituntil.c#L221

• if nelems > 0 and all entries of status are nonzero, then the routine it will loop forever and deadlock.
• With only one of the ivars entries passing the test, eventually you will get n == nelems and the busy loop finishes.

[jdinan]

There is an erratum to OpenSHMEM 1.5 for this -- https://github.com/openshmem-org/specification/wiki/OpenSHMEM-1.5-Errata