Filters at the end works faster than at the beginning

[FrancescAlted]

I have detected this issue while doing some benchmarking, and I 'fixed' the performance by putting the filter at the end of the pipeline. I suspect that this can be optimized in the forward pipeline of C-Blosc2 (check whether the backward pipeline is affected too).