AVIF export: write HDR10 content light level (MaxCLL/MaxFALL) for PQ

[MaykThewessen]

PR: AVIF export — write HDR10 content light level (MaxCLL/MaxFALL) for PQ

Branch: upstream-avif-hdr10-clli (1 commit, src/imageio/format/avif.c, +56 / -0)
Base: darktable-org/darktable:master
Refs: part of #18078

---

What this does

darktable already tags AVIF exports with the correct nclx color information
(colorPrimaries / transferCharacteristics / matrixCoefficients) for PQ and
HLG Rec.2020 and P3 output profiles. However, it never writes the HDR10
content light level box (clli: MaxCLL and MaxFALL). Many players, OSes and
HDR displays use those values to tone-map, and several platforms (e.g.
iOS/Instagram) expect them on HDR stills.

This PR computes and writes clli for PQ output.

How

For PQ (SMPTE ST 2084) the exported float samples are absolute-luminance
encoded, so the real content light levels can be derived from the pixels via the
PQ EOTF:

• MaxCLL = the brightest single sample (max RGB component), in cd/m².
• MaxFALL = the mean over all samples of that per-sample peak light level
  (the frame-average light level; for a still, the whole image is the frame).

These are written into image->clli (maxCLL / maxPALL). A debug line is
emitted under DT_DEBUG_IMAGEIO.

HLG is not handled here: it is a relative system with no fixed nit scale, so
clli is not meaningful and is left unset.

Scope / notes

• Only active when the output is PQ and nclx is being written natively.
• No new dependencies; the PQ EOTF is a small static helper.
• The computation reuses the same (const float *)in, 4-channel layout the rest
  of write_image() already uses.

Testing

• Built on macOS (Apple Silicon) against libavif.
• Export an image through a PQ Rec.2020 output profile and inspect the file,
  e.g. avifdec --info out.avif or exiftool out.avif, and confirm the
  MaxCLL / MaxFALL values are present and sane (debug log prints the values when
  run with -d imageio).

[MaykThewessen]

Tested end-to-end

Exported a test image through the PQ Rec.2020 output profile to 12-bit AVIF
(darktable-cli ... --icc-type PQ_REC2020) on macOS / Apple Silicon, and verified
the bitstream with two independent tools.

avifdec --info:

Bit Depth      : 12
Color Primaries: 9     (BT.2020)
Transfer Char. : 16    (SMPTE ST 2084 / PQ)
Matrix Coeffs. : 9     (BT.2020 non-constant luminance)
CLLI           : 8970, 4615   (MaxCLL / MaxFALL, nits)

ffprobe:

color_primaries=bt2020
color_transfer=smpte2084
color_space=bt2020nc
pix_fmt=yuv422p12le
Content light level metadata: max_content=8970, max_average=4615

With this PR the clli box is present and the values track the actual pixel
luminance via the PQ EOTF; without it the box was absent. (The high numbers here
are because the synthetic test image is fully saturated with no filmic roll-off;
a normally graded photo lands much lower.)