Change texture_mean to work for all mosaic pattern widths

[Alex-Vasile]

The current implementation works fine when a single mean in required, but it did not function when a per_subpixel mean was required. A per-sub pixel mean is required for hot pixel identification and correction. This version is generalized and should work for all mosaic patterns.

How this approach works:

1. Calculate the per-subpixel sum along rows.
   • Starting with an row x col sized texture, the output is a MPW x col texture.
2. Sum the per-subpixel along columns.
   • Starting with the MPW x col image the result is a MPW^2 length buffer
3. If needed, sum to single value, going from MPW^2 values to a single value.
4. Divide sum by number of pixels.
   • per_sub_pixel=false: length * width
   • per_sub_pixels=true: length * width / mosaic_pattern_width^2

Both Bayer and XTrans images are affected by this, I tested both. Image results are identical before and after this PR