Provide a more efficient process for float upsampling ratio

[savmickael]

No description provided.

[dumasl]

So far the upsampling strategy when ratio is p/q real value is decomposed in two steps:

• zoom by a factor q
• decimate by a factor p

We could manage to zoom by a factor p/q in one single step but we need to be careful with block decomposition so that what we compute lies correctly on the output grid. Keep in mind that the first output computed pixel is linked to the first input pixel.