[LangRef] make per-element poison behavior explicit

As discussed in D70246 and PR43958:
https://bugs.llvm.org/show_bug.cgi?id=43958

The LangRef seems ambiguous about the behavior of poison with respect
to vectors.

We could go further with text and/or examples - suggestions welcome.

Also, see discussion on llvm-dev;
http://lists.llvm.org/pipermail/llvm-dev/2019-November/137243.html

Differential Revision: https://reviews.llvm.org/D70641