fix(text): activate verticalAlign on maxHeight alone

[chiefcll]

Summary

• verticalAlign was silently ignored unless contain included height. Setting verticalAlign: 'middle' on a node with only maxHeight did nothing. This splits the gating so verticalAlign activates whenever maxHeight > 0, while the mountY-against-maxHeight override stays tied to contain & height (preserves existing mount semantics).
• Removes the stale "Not yet implemented in the SDF renderer" docstring on the verticalAlign prop — CoreTextNode already drives the offset for both backends — and corrects the documented @default (Stage.ts resolves missing values to 'top', not 'middle').
• Adds a 3-line row to the visual regression matrix (taller 280×280 container) to catch any first-line vs last-line asymmetry across top / middle / bottom.

Decisions

• 'top' semantics: kept as CSS line-box (current behavior — half-leading above the first cap-top). No new 'top-tight' enum.
• 'bottom' semantics: kept as CSS — half-leading slack below the last descender.
• 'baseline': explicitly considered and skipped (no obvious meaning in a block-vs-maxHeight context).
• Activation: maxHeight > 0 alone. This is the only behavior change.

Test plan

• pnpm build clean.
• vitest run — 168/168 pass.
• Full VRT in Docker CI mode — 163/163 pass against the new certified set (only text-vertical-align-{1,2}.png changed; all other text/scaling/contain snapshots produce identical bytes because the default is unchanged).
• Verified rendered output for 1-line / 2-line / 3-line × top / middle / bottom × sdf / canvas matches expectations (cap-anchored baselines land on the green reference lines).

🤖 Generated with Claude Code