Add no_aos_sampling GALLIVM_PERF option

This forces using general sampling and should improve precision and performance in some cases.

As discussed on the mailing list, in some cases when sampling from 8 bit textures with llvmpipe, the interpolation result differs significantly from results obtained using hardware texture sampling. As explained by Roland Scheidegger, this is due to performance optimizations in llvmpipe which perform the sampling with 8 bit precision. This patch enables the quick fix (also suggested by Roland Scheidegger, which activates floating point sampling by forcing the general code path) if the option no_aos_sampling is passed via the GALLIVM_PERF environment variable.