[ARM] Add neon FP16 scalar_to_vector patterns.

This adds some simple fp16 scalar_to_vector patterns, preventing a
selection failure if this came up.

Differential Revision: https://reviews.llvm.org/D95427