[prakashUXtech]

Part of #436 (retouch-heal RH-5, slice RH-dab).

What

Builds dodge, burn and sponge in Sources/CoreRaster/Heal/ToneDabs.swift, keeping the RH-0 signatures:

• A dab is placed with its top-left at (x, y). Dab cells that fall off the buffer are skipped, and any Int position is safe because placement is checked before any offset is added. Only pixels with coverage above 0 and alpha above 0 change. Alpha never changes.
• Dodge scales linear colour by 1 + exposure × coverage × rangeWeight(luma). Burn scales it by 1 − 0.5 × exposure × coverage × rangeWeight(luma). Sponge scales chroma about Rec. 709 luma by 1 + amount × coverage. Colour is clamped to 0...1 straight (0...alpha premultiplied). Scaling commutes with premultiplication, so the ops work on premultiplied values without converting alpha back and forth.
• Strengths outside 0...1 (dodge and burn) or -1...1 (sponge), or NaN or infinite, throw RasterError.invalid. A strength of 0 returns at once, so it is an exact identity.

Why the midtone curve differs from the packet

The packet's pseudocode uses a midtone weight of 1 − |2l − 1|. That weight is 0.2 at luma 0.1 and 0.1 at luma 0.95, which breaks the Acceptance table's "within 1/255". As the lead decided, the Acceptance table wins. Midtones is now a smoothstep window, smoothstep(0.2, 0.4, l) × (1 − smoothstep(0.6, 0.8, l)): it is 1 over 0.4...0.6 and exactly 0 at or below 0.2 and at or above 0.8. Shadows (1 − smoothstep(0.2, 0.4, l)) and highlights (smoothstep(0.6, 0.8, l)) mirror it, and the three weights add up to 1. The curve is documented on TonalRange.weight(luma:). Because the midtone weight is exactly 0 below 0.2, the RH-7 walk's "30-dab stroke leaves luma under 0.2 within 1/255" also holds.

Tests

• ToneDabTests is turned on, all 8 RH-1 cases: a midtone dodge lifts luma 0.5 and leaves 0.1 and 0.95 within 1/255; burn mirrors it; sponge +1 raises chroma and −1 greys; coverage weights the change; alpha never changes; exposure 0 is identity; nothing outside the dab changes; fuzzed strengths and positions either throw RasterError or dab. I also added shadowsAndHighlightsOnly, which checks that the mirrored ranges move only their own patch.
• RetouchContractTests.dabStubsThrowRasterError: removed. With RH-6 (feat(raster): smudge dabs drag a carried colour that fades by strength #602) having removed the smudge check, only the dodge, burn and sponge stub checks were left, so the whole test goes, as that file's header says each task does. This file is outside the task's file list, but scripts/ci cannot pass without the change.
• Mutations in Tests/mutations/CoreRaster.ToneDabs.json: the inverted midtone weight is caught by ToneDabTests, and so is ignoring coverage (by dabCoverageWeights).

C4: Sources/CoreRaster/Heal/ is already owned by the image-ops component. No platform or library API is used beyond the Swift stdlib's min, max and ClosedRange, so I did not consult the docs tools.

Belt-Lead: act-1a150b99ea1-n4z4
Feature: retouch-heal
Slice: RH-dab
Mutate: 2/2 CoreRaster.ToneDabs
Fast-Check: pass
Head: 30fb494

[github-actions]

Fast lane: merged into dev (142 lines, 4 files, Fast-Check marker).