arrayfire imax function

`max_val, max_idx = af.imax(img)`

running this function seems to have wrong behavior and finds wrong max value for an image with pixels in range [0,1].

When running `max_val, max_idx = af.imax(af.flat(img))` the correct max value is found.

Reproduce the discrepancy with an image whose pixels are in [0,1], comparing af.imax(img) with af.imax(af.flat(img)). Start at the Python binding entry point for af.imax and inspect how image shape is handled; done means both forms return the same correct maximum value and index.

Written by the indexing model from the issue text.