`irfft` returns `float64` for an input with dtype of `float32`

`irfft` returns `float64` for an input with dtype of `float32`

```python
import mkl_fft, numpy
a = numpy.ones(10, dtype="f4")

mkl_fft.irfft(a).dtype
# dtype('float64')

numpy.__version__
# '2.2.3'
numpy.fft.irfft(a).dtype
# dtype('float32')
```