fixed-point: changed find_spectral_pitch() to use single-precision (16-bit) FFT.

This involved adding kfft_single.[ch] that redefines kiss_fft a second time
with a different prefix. All this is still a bit of a mess now. The mask
had to be converted to 16-bit input, but we're still using floats to apply it.

libcelt/pitch.h

@@ -41,6 +41,9 @@
 #include "kiss_fftr.h"
 #include "psy.h"
 
+kiss_fftr_cfg pitch_state_alloc(int max_lag);
+void pitch_state_free(kiss_fftr_cfg st);
+
 /** Find the optimal delay for the pitch prediction. Computation is
     done in the frequency domain, both to save time and to make it
     easier to apply psychoacoustic weighting */