Add tests for folding QuantizeToF16 with inf

We had tests that were suppose to test fold QuantizeToF16 instruction with both positive and negative inf; however, they were code incorrectly. They relied on two instructions being folded, when our folding tests only fold 1 instructions. They will be removed, but new ones should be added.

Locate the existing folding tests for the QuantizeToF16 instruction and inspect how they fold a single instruction. Remove the incorrectly structured cases, then add separate tests for positive and negative infinity that each exercise one QuantizeToF16 instruction; the tests should pass using the existing folding-test conventions.

Written by the indexing model from the issue text.