test(mindmap): #204 pin HLSColor lighter/darker readability contract (20 tests)

[jsboige]

What

Pin the mind-map node readability color contract for HLSColor.GetLighterColor(string) and GetDarkerColor(string) — pure string→string color transforms with zero coverage, used in 6 production call sites.

Why

The mind-map builder derives two shades from each fallacy family's color:

• GetLighterColor(string) → node.BACKGROUND_COLOR (a lighter tint = the node fill)
• GetDarkerColor(string) → node.COLOR (a darker shade = the node text)

Both are called from SetNodeStyle in MindMapDocumentConfig, FallacyMindMapDocumentConfig, and VirtueMindMapDocumentConfig (6 call sites total). Together they enforce the readability invariant: on a family-colored node, the text is darker and the fill is lighter than the raw family color.

A regression that flipped a direction — darker becoming lighter than the base, or lighter becoming darker — would make every node's text unreadable (dark-on-dark / light-on-light). Because mind-map rendering is currently gated behind visual QA the cluster cannot always run (the #565 pixel-RTL wall), this is exactly the silent-aesthetics regression a contract test must catch in isolation.

Tests added (20)

Pure & deterministic; additive only (no production code changed). Luminance proxy = Rec.601 weights (0.299R+0.587G+0.114B), a stable viewer-independent direction measure:

1. Darker is darker — GetDarkerColor yields lower luminance than base, across 5 real family colors (#811da3 purple, #555555 gray, #08af93 turquoise, #8dc801 green, #dc0f0a red) — the text half
2. Lighter is lighter — GetLighterColor yields higher luminance than base, same 5 colors — the fill half
3. Both return valid parseable hex — the .mm node holds the string verbatim into the XML; an invalid value corrupts the file
4. Determinism — same family color → same shade (builder calls once per node)
5. Transforms change the color — anti no-op regression (output ≠ input for a mid-range color)
6. Lighter is lighter than darker for the same base — the relative ordering the light-fill/dark-text scheme assumes
7. Characterization — exact observed outputs for purple #811da3 (lighter #CB76E7 / darker #450F57), captured from the real ColorTranslator HLS round-trip — catches a vendored-HLS refactor

Verification

• dotnet test --filter HlsColorContractTests → 20/20 pass
• Full suite → 506/0/5 (baseline 486 + 20). 0 regression.
• Build: 0 error (only preexisting nullable warnings).

Verify-before-code note

ColorExtensions.GetContrast (+ its helpers ConvertToHSB/ConvertToRGB) in the same file are dead code (defined, never called in production — grep confirms zero call sites outside the file). Excluded from this PR; flagging for a separate cleanup if desired. This PR targets only the live GetLighterColor/GetDarkerColor path.

Context

Cluster is degraded (po-2023 down, DNN/régén orphaned). #204 pure-helpers lane is closed; this PR extends coverage into the mind-map rendering zone — relevant given #565 (8-lang régén) is pending and #568 (mindmap reliability) is open. The readability invariant is precisely the contract a worker can pin while visual QA stays ai-01-only.