Implement convolve8_avg using common functions which are already optimized
instead of using more obscure ones which have only C versions. Encoder
overall speed-up of about 12%.

Change-Id: I8c57aa76936c8a48f22b115f19f61d9f2ae1e4b6