[RISCV] Add support for half-precision floats

Complete fp16 support by ensuring that load extension / truncate store
operations are properly expanded.

Reviewers: asb, lenary
Reviewed By: lenary
Differential Revision: https://reviews.llvm.org/D69246