Move Dye Mute into density space; rename Lab Saturation to Chroma

[marcinz606]

Follows #664, which added density-space Print Saturation.

Why

Dye Mute was never an operator. It multiplied a grade-coupled scalar into Lab Saturation:

slopes = context.metrics.get("print_slopes")
damp = 1.0 if slopes is None else grade_chroma_damping(slopes[1], self.config.chroma_damping)
eff_sat = self.config.saturation * damp

One scalar on a CIELAB a*/b* scale, folded identically on the GPU. So its physical justification lived in the exposure stage (the green print-curve slope) while its effect landed post-decode in a perceptual space with no relationship to how the paper produces colour.

Print Saturation corrects the same phenomenon — a harder grade steepens each channel's slope, widening per-channel density spread and inflating chroma — in the domain where it originates. migrations.py already recorded that the density-space damping prototypes were dropped as "mathematically redundant with density_saturation itself", but the Lab-space one kept shipping. Worse, the two couldn't see each other: print_slopes is computed from grade/density/trims only, so it never observed density_saturation. Pull both and you got uncoordinated double correction.

What changed

Dye Mute now damps the global density_saturation:

k = density_saturation * (slope_min / slope_g) ** strength

Per-layer trims sit outside the damping — a crossover correction is a measured fix for one paper's diverging dye layers and must not shrink because the grade hardened.

Folded in at the processor on both paths. slopes is already available before the curve call on CPU (exposure/processor.py:233) and GPU (gpu_engine.py:1172), so nothing new crosses into the kernel or a shader: no WGSL change, no new buffer, no bind-group arity change.

LabConfig.chroma_damping → ExposureConfig.density_saturation_damping. Moving between sub-configs costs no migration at all — from_flat_dict routes flat keys by which dataclass declares the field.

Default is 0.0 (off), was 0.25 (on)

grade_saturation_damping(s, 0.0) returns exactly 1.0 for every slope, so resolve_saturation_matrix still returns None at default and #664's byte-exact default path is preserved — the 3×3 matmul stays off unless the user opts in.

The old value is dropped, not carried

One DROPPED_KEYS entry rather than a KEY_RENAMES line. 0.25 damping on CIELAB chroma and 0.25 damping on density separation are not the same effect magnitude — the density version self-tapers at toe and shoulder — so carrying the number would silently apply something the user never set. There is no honest 1:1 remap between the domains and a fudge factor would only hide that.

This means anyone who deliberately cranked Dye Mute loses that setting and needs to re-dial it, in the domain where the number now means something.

Lab's "Saturation" is now "Chroma"

Two sliders named Saturation in two tabs was the exact ambiguity this change exists to remove. Chroma names what it does — an even CIELAB a*/b* scale after decode. Field stays saturation; label, catalog row, shortcut description and tooltips updated. Both controls' tooltips now explain the contrast and point at each other.

Also closes gaps left by #664

Print Saturation shipped with no shortcut_registry entry, no settings_catalog row, no slider_shortcut_groups entry and no tutorial mention. Its slider also stayed visible and live in B&W while the processor forced it to identity — a control that did nothing. Both fixed, and Dye Mute's four existing entries moved Lab → Exposure.

Shortcut action ids and the slider group id are renamed to track the field. That discards any custom binding or step size for these two actions (both persist keyed by id, and unknown keys are dropped on merge). Accepted deliberately: they ship with no default key, and a permanent name mismatch across five files is the worse cost.

Look change

Default-config chroma rises, because Lab no longer carries the damping multiplier and Dye Mute now defaults off.

Verified as the only delta before regenerating the golden: lab.saturation = 0.9210 reproduces the old values to 2.7e-4 against a 1e-3 tolerance, and that number is exactly grade_saturation_damping(2.780, 0.25). The large encoded swings in _GOLDEN (0.249 → 0.084) are the Adobe TRC amplifying small linear changes near black on the saturated test patches. test_characteristic_curve stayed green throughout, corroborating that the exposure kernel genuinely didn't move.

Tests

• test_chroma_damping.py → test_saturation_damping.py. Pure-function tests survive under the new name; added that strength 0 is exactly 1.0 for every slope (the guarantee the default rests on).
• Lab Saturation is now grade-independent — renders identically whatever print_slopes says, including when absent.
• Damping reduces chroma at a hard grade, and the effect grows as the grade hardens — the reason Dye Mute survives as its own control instead of folding into a static value.
• Trims stay outside the damping; B&W is inert.
• A legacy chroma_damping key loads with no unknown-key warning and does not populate the new field.
• test_gpu_curve_parity.py::test_cpu_gpu_match_dye_mute — CPU/GPU parity at non-default damping against a real paper dye matrix, since the two paths compute the factor separately and could drift.

2672 passed, 1 skipped. Lint and type clean; ruff format left all 456 files unchanged.

Docs: PIPELINE.md (§3 Dye Mute, §6 Global Saturation no longer grade-coupled) and USER_GUIDE.md (Print Saturation + Dye Mute under Tone, Saturation → Chroma, cross-referenced both ways).