attribute chips: fix exclusion-group width stretch + add group boundary (#742, #743)

[WilfordGrimley]

Summary

• Fixes What's That: exclusion-group chip rows stretch to the full panel width, with the Yes/No controls stranded at the far right #742: an exclusion-group chip (Black/White/Silver Border, Old/Modern
  Border, Future Frame) rendered at the panel's full column width - measured
  494px on an 800px viewport, ~1170px at 1600px - with the Yes/No controls
  stranded far from their label. Root cause confirmed live in a real browser:
  once the chip's row switches to flex-direction: column, the flex spec's
  default align-items: stretch stretches any child with no definite
  cross-size to the column's full width. ChipGroup now sets
  width: fit-content, which is immune to that stretch regardless of which
  axis the ancestor row happens to run on - a one-line fix that doesn't
  require touching the existing breakpoint.
• Fixes What's That: the attribute chip panel renders two independent exclusion groups as one unlabelled list #743: BORDER_COLOR_GROUP and FRAME_STYLE_GROUP are independent
  axes, but nothing rendered their existing label field - the two groups'
  chip rows ran together as one unlabelled list, so a correct half-collapse
  (one group's siblings dimming, the other untouched) read as a bug. Each
  group now renders its label as a heading, plus a divider between them in
  the flat-stack (Level 2) layout - the ring layout keeps them apart
  spatially already, so it doesn't need one.
• Amends docs/proposals/mockups/wtc-rebuild/SPEC-wtc-rebuild.md (A14) to
  record both fixes, following the file's existing amendment format.

Collapse behaviour (which chips dim or hide when a sibling is selected) is
unchanged - only its legibility was at fault.

Test plan

• npm test - 73 suites / 676 tests passed
• npx playwright test - 345 passed, 6 skipped (pre-existing flakes,
  unrelated to this change), 0 failed
• npm run build - compiled successfully
• python3 .github/scripts/coverage_delta.py --base origin/master - clean
• Measured live in a real browser (Chromium, via Playwright) before and
  after: exclusion-group chip width at 800px went from 468px (stretched
  to the panel) to 162px (content-sized, matching the standalone chips'
  density); at 1600px from 1164px to 162px; at 390px it was already
  162px both before and after (below the 576px breakpoint that
  triggers the stretch)
• Confirmed the chip panel never intersects the pinned reference card's
  bounding box at 390px, 800px, and 1600px
• Confirmed "Border Color" / "Frame Style" headings render and the
  collapse-siblings-not-vote test still passes unchanged