Add Kodak paper profiles fitted to datasheets

[marcinz606]

Summary

Five new paper profiles, fitted to Kodak datasheet characteristic curves. The curves were read exactly from the PDFs' vector paths, not by eye.

Profile	Mode	Source	Fit RMS
Kodak Polymax II RC	B&W	G-26, 7 filter grades	0.022 D
Kodak Polycontrast IV RC	B&W	G-4037, 4 filter grades	0.039 D
Kodak Portra Endura	RA4	E-4021, Status A R/G/B	0.030 D
Kodak Supra Endura	RA4	E-4021, Status A R/G/B	0.036 D
Kodak Ultra Endura	RA4	E-4020, Status A R/G/B	0.043 D

Existing profiles are byte-identical. The Endura dye matrix moved into a shared constant with the same values.

Fit method

scripts/fit_paper_profile.py is rewritten. It renders through the app's own print_curve (default sliders, paper white on, no BPC). It does not use a copy of the curve math.

The slope is not free. grade_to_slope sets k = 2.9 / (R/100) density per log E, independent of the negative, so each curve's slope comes from its measured ISO R. With a free slope (the old script), the fit converged on soft knees (sharpness ≈ 1) under a steep line. That shape renders flat at the user's grade. d_min is the measured base. One shape is shared across a B&W grade family. For RA4, the shape is fitted on the mean curve, then each channel gets a slope gain (channel_gamma) and a base tint.

Constraints found

• Midtone gamma headroom. Snap (±0.5) and its per-layer trims (±0.5) add to the profile's paper_midtone_gamma, and the kernel needs |γ| < 1. Unbounded fits wanted 0.5–0.63, so the fit bounds it at ±0.25. A new test holds every profile to that.
• Dye matrix. Matrices integrated from the datasheet dye-density curves at Status A kept only 42–75% of chroma, against 76–89% for the existing papers, so prints looked dull. They hold the dyes' unwanted absorptions but not the interimage effects that offset them. The E-4021 curves also carry a ~0.15 D baseline. The new RA4 papers share the existing Endura matrix.
• Per-channel D-max spread (Premier red 2.76 vs blue 2.44) cannot be expressed: the profile has one d_max.

Not included

• Fuji papers: the datasheets have dye-density curves only, no characteristic curves.
• Foma Fomaspeed Variant III: the PDF draws its curves as fragmented outlines that cannot be split by grade.
• Ilford MG RC, MG FB Warmtone, Kentmere VC Select: raster curves only.
• Endura Premier (E-4070, RMS 0.031) and MG FB Classic (RMS 0.034) also fit their datasheets, but the profiles keep their hand-tuned values. A switch would change existing edits and needs a THUMBNAIL_RENDER_VERSION bump.

Test plan

• make all: 8750 passed
• Headless app run (Xvfb): each new paper appears in the Paper dropdown for its mode, renders finite pixels and differs from Neutral
• Existing 7 profiles field-identical to main
• Visual check of the new papers on real scans