Share one dialog shell across main-window dialogs

[enaboapps]

Closes #889. Stacked on #898, tracked in #891.

What changed

• Setup, pairing, profile, confirmation, switch practice and capture dialogs share one surface, a 16px radius, 28px padding (20px under 480px) and one shadow. Headers use one title size. Footers share a divider and spacing, and their spacer pushes the main actions to the end.
• The setup guide was fixed at up to 900px tall, so short steps floated in empty space. It now sizes to its step, with a 480px minimum so it doesn't jump between steps, up to the window height. Only the middle scrolls.
• The setup guide's label is sentence case instead of an uppercase eyebrow, and the step rail is thicker.
• Each pairing request is a card. The verification code is set large on its own surface for comparison, and Reject and Accept sit at the end.
• Opening a dialog fades and rises over 160ms. prefers-reduced-motion turns this off, along with the capture pulse and button transitions.
• Dialog containers show a focus ring only for keyboard focus. Focus trapping, inert backgrounds and the capture dialog's key swallowing are unchanged.

Validation

On this head:

• npm run lint passed
• npm test passed: 238 Vitest tests and 5 Node tests
• npm run build passed
• cargo fmt --check, cargo clippy --all-targets -D warnings and cargo test passed (566 tests, 1 ignored). No Rust changes in this stack.
• Rendered the setup guide, profile editor, pairing request and removal confirmation in headless Chrome at 920px and 620px, light and dark. Pairing and confirmation used a mocked Tauri bridge.

Review follow-up

An independent review of this layer's head raised these points, now addressed:

• Long device names without spaces wrap in the pairing card instead of running under the code. A comment records why the entrance animation must never leave a transform on the setup guide: the capture dialog renders inside it.

🤖 Generated with Claude Code

Rebase validation (2026-09-25)

Rebased onto current main (7fb8fce, including #881) in stack order. git range-diff shows every commit in this PR is patch-equivalent to its previous head, and the intended parent remains its PR base. On the final stack tip (40c75cc), Node 24.19.0 lint, tests and build passed; Rust 1.97.1 format check, Clippy and tests passed. An independent review of the updated PR heads found no actionable findings. CI is rerunning on each new head.