af_biquad: avoid subtraction in loop

17% faster
Note the speedup is smaller if its done before
storing in the context (gcc 4.6.3 core i7)

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>