docs(design): keep DESIGN.md to rules that stay true

[Astro-Han]

Summary

DESIGN.md is read by agents as much as by people, so a stale line is worse than a missing one — it gets adopted as fact. This removes the two kinds of line that cannot stay true, and fixes the assertions that were already wrong. Doc-only: no token or product CSS changes. 248 → 238 lines.

1. Prose that tracks the migration's current state

Removed from §3, §4, §5 and §11:

• the One Colorspace exception, "until T4 unified it"
• --ring-soft, "was called --shadow-minimal-flat until T4"
• the Floating Recipe, "currently enforced … a transitional implementation"
• the wash ban, "T4 migrated fourteen such sites … the rest are debt"
• the link rule, "--link currently resolves to raw --accent"

This branch is its own evidence for why. An earlier revision of it documented the srgb exception, the --shadow-minimal-flat rename and the unenforced Floating Recipe as open items. #2536 closed all three the same day, and every one of those lines had to be rewritten before this could land. A spec that tracks progress has to be rebased against progress.

The rules are unchanged and read stronger without the timestamps: the ladder derives in one space, --ring-soft is a border, the recipe binds every portal including Astryx's own. Where a rule and the code genuinely disagree, that belongs in an issue rather than in a clause hedging the rule.

2. Values and inventories that source already owns

• The surface pixel table and the ink contrast columns. Their caption claimed the contract tests held them; ink-ladder-contract pins two structural invariants and no number.
• The seven-alias legacy surface list, which this file's own last Don't forbids copying in. Its one load-bearing fact — --background is the card fill, not the page color — moves into §2's opening sentence.

Frontmatter stays. It is the machine-readable half of the DESIGN.md convention and its consumer is the agent reading it, not a build script, so a wrong entry gets corrected rather than deleted.

3. Assertions that were wrong

Where	Was	Is
§6	4px inner and 16px page tiers	Neither exists — --radius-inner resolves to 6px, --radius-page to 28px. Removed from table and frontmatter
§6	—	The relationship those rows hid: control/card/container are the same numbers as Astryx inner/element/container. New Astryx column + Two-Name Rule, which notes the pairs agree today as independent literals and can still be split upstream
§8	"there is no info status semantic"	Fused the color families with the dot semantics. --info is live in eleven palettes; what has no info is StatusSemantic. Split across §8 and §9
§8/§9	—	Astryx Badge and StatusDot carry fixed neutral-theme literals, so an Astryx info pill and Maka's --info are unrelated blues
§3	—	Astryx's Text color="secondary" reads --color-text-secondary — a fixed literal outside the bridged containers, --muted-foreground inside. Three results for one word, live call sites on both sides
§1	maka-tokens.css owns "the Astryx bridge"	It owns one third; makaTheme.ts and astryx-mount.css own the rest
§7	Heading 1–5	heading-6 and large are real roles and were missing
§4/§9	--border-strong "no fills" vs scrollbar thumb using it as a fill	Reconciled on §9's side, since T5 shipped the thumb

Verification

• npm run format:check green; diff touches DESIGN.md only.
• Every claim checked against source: expandRadiusScale.js, generated maka.css, astryx.css atomic classes, maka-tokens.css, astryx-mount.css, packages/ui/src/status-vocabulary.ts.
• Rebased onto current main (post-T5).

Related: #2525 (T1), #2536 (T2–T4), #2538 (T5), task #168.