SimplificationPass should fold FConvert of float32 constant to float16

ConvertToHalfPass will generate the following:

%float_0_5 = OpConstant %float 0.5
...
%156 = OpFConvert %half %float_0_5

It would be nice if SimplificationPass would fold this into

%half_0_5 = OpConstant %half 0.5

Attention @s-perron

Start by locating SimplificationPass and ConvertToHalfPass, then inspect existing constant-folding tests for floating-point conversions. Done means the shown float32 constant conversion is folded into the equivalent float16 constant, with coverage in the relevant test suite.

Written by the indexing model from the issue text.