Generate single-precision floats when they are equivalent to their double-precision representation, even when not in canonical mode. Always generate singles for Float32Array, and always generate doubles for Float64Array.
Coverage remained the same at 100.0% when pulling ee4ab054521d0e46817d059a6e39286e4b138028 on fround into 474fb2bd7328cba3e9c9b416d5248f809983d07e on master.
Generate single-precision floats when they are equivalent to their double-precision representation, even when not in canonical mode. Always generate singles for Float32Array, and always generate doubles for Float64Array.