Speed improvement
Added support for large dataset/scale processing in reasonable time.
Example: The time to produce the CWT of a 140k sample with a scale range [1-200] goes down from 1min51s to 14s on a 2.9Gh Core2 Duo
Replaced pywt.cwt() by a drop-in replacement with adaptative convolution algorithm selection,
the scale processing loop switch to FFT based convolution when the complexity is better in N*log2(N)