test for fp16 sqrt is not using a correctly rounded reference

[lakshmih]

[shajder]

@lakshmih similar situation to #1996, I compared all reference values with equivalents computed with single precision - there are no differences but still I've created the PR to use single precision reference computations

[bashbaug]

Is there more to do here, or can this be closed?

[bashbaug]

Closing as discussed in the June 3rd teleconference.