While trying to snap to the closest 3x3 pattern for xtrans sensors we would have to expand
the roi_in size, if not doing that (code before commit) we might miss the last column/ror
resulting in black pixels in cpu code.

Unfortunately the OpenCL code for xtrans demosaicers has issues in this situation so we
don't snap to closest but to closest-top/left.

Also tested tiling.

Motivated by #14888 to fix similar issue found in retouch.

Fixes #14888.

Fixes https://github.com/darktable-org/darktable/issues/14717

While investigating issues reported in https://github.com/darktable-org/rawspeed/pull/479#issuecomment-1632595404
also checked demosaicer code. Two issues have been identified and fixed in this pr:

1. The lmmse demosaicer didn't have a border specific code, as we process raw-prepare cropped data in roi instead
   of sensor data which is cropped after demosaicing we have to do this as in the other demosaicers.

2. The default RCD demosaicer can have interpolation issues in the outermost 7 (instead of 6) rows/colums under
   certain conditions.
   While being here re-tested scheduling on todays compiler/omp, no reason to take the dynamic scheduling any more.

As reported and discussed in https://github.com/darktable-org/rawspeed/pull/479#issuecomment-1632595404
there is an issue in the cpu code of temperature module for bayer sensors.

Fix the omission of rounding-up for colorout and legacy colorbalance
corresponding to that found in #14886, which fixes a failure to
process the last few pixels in the last row of the image under certain
circumstances.

For some combinations of roi_out width&height and darkroom canvas the calculation of
chunksize is too small leading to some pixels at the lower-right border to be left
out from calculation.

This pr chose one way to fix, possibly increase chunksize by 4 in such case.
The alternative would have been to check for real-work for the thread with highest id
which would be slower.

Fixing usage of a wrong module string.

Refine previous fix as we don't want to add the highlights module
for non RAW files.

Continued work for #14050.

This part in the introspection circuitry was missing.

This second version keep compatibility with old edits by adding
an highlights module with CLIP method when no highlights module
found in the XMP.

Fixes #14050.

This reverts commit 37e1a8ba.

This reverts commit 54689ece.

This part in the introspection circuitry was missing.

Fixes #14050.

Previously it only worked for mm. Fixes #14855.

We have to take care of channels while allocating temporary buffers in OpenCL
blending code.

Fixes #14844.  Does not address the long-standing unintuitive behavior
that negative EVs brighten the portion of the image which positive EVs
leave unchanged.

When checking if a default auto-preset is to be used properly
check that user's preset do apply to the current image. That
is for the camera/lens/iso/exposure/aperture of the image.

Fixes #14799.