Handle Real-valued Circulant and Toeplitz

[ChristophWWagner]

Get rid of extending the data type to complex when dealing with real-valued transforms. Any idea on how to do this, anyone?

I'd give reworking to real-valued DFTs internally a shot and would then zip the real and imaginary portions of the input and output vectors accordingly in the complex case. This way we'd only construct one transform data structure at a time.