Display color spaces should mirror rather than clamp negatives

[doug-walker]

The current display color spaces that use ExponentTransform are using the Basic style of negative value handling, which clamps them. In video workflows, it is desirable to preserve negative values so that these may become extended-range undershoot values (e.g., values below 64 in a 10-bit SMPTE/narrow range signal).

The proposal is that these displays be modified to use the Mirror style of negative values handling, which reflects them around the origin. These entries in the config use built-in transforms, so the proposal is to add new built-in displays that use the different negative style and then update the configs to use them.

Note that in OCIO 2.4, the built-in PQ and HLG implementations were changed to mirror negatives, so those are already fixed.

For displays that use the ExponentWithLinearTransform, those are doing linear extrapolation for negatives, so these are ok. However, they could be updated to use mirroring, for consistency, if desired.

[scoopxyz]

Made this comment in the TSC/WG. My initial thought is to add the mirror behavior to the built-in-transforms but add an additional range transform to clamp negatives in the ACES configs. This would be to preserve existing behavior. If a user really desired to have negatives preserved, and sensibly handled, they could feasibly delete that transform manually (acknowledging that many config authors directly edit the YAML).

[doug-walker]

I've made a PR to add mirrored built-in transforms to the library:
AcademySoftwareFoundation/OpenColorIO#2183

[doug-walker]

This has been completed in the current draft configs, the following color spaces are using the new mirrored styles:

• sRGB - Display
• G2.2 Rec.709 - Display
• P3-D65 - Display
• Rec.1886 Rec.709 - Display

All the other display color spaces were already using mirroring.

[doug-walker]

Implemented in PR #160.