Modernize UI/UX (tracking: post-Tailwind-v4 work)

[srid]

Tracking issue for the visual + UX modernization work that follows the Tailwind v4 migration. Split out deliberately so the v4 migration lands and stabilizes in isolation — regressions there should be attributable to the tokens change, not bundled with redesign.

1. Visual refresh

• Spacing, typography, sidebar density pass on the existing layouts in emanote/default/templates/.
• Native <details>-based collapsible sidebar, replacing the toggleHidden() inline script in components/breadcrumbs.tpl:41-45 and the Heist-conditional expand/collapse in components/sidebar-tree.tpl:83 (existing TODO). This is genuine JS elimination.
• TOC scroll tracking (components/toc.tpl:21-73, existing TODO "We should get rid of JavaScript!" — issue Render TOC without JavaScript #520): swap window.onscroll + offsetTop arithmetic for IntersectionObserver. This improves the JS, it doesn't eliminate it — see section 3 below for the module boundary.
• Proper mobile sidebar drawer (overlay, scroll-lock, focus trap). Current mobile UX is a class-toggle, not really a drawer.
• Icon partial: consolidate the copy-pasted inline heroicon SVGs in sidebar.tpl, breadcrumbs.tpl, toc.tpl, etc. into a single reusable Heist partial.

4. UX features

• Command palette (Ctrl+K already opens Stork search; extend to actions: jump-to-tag, jump-to-recent, toggle sidebar, toggle theme).
• Keyboard navigation through the sidebar route tree (arrow keys, Enter to open).
• Recently-viewed notes (localStorage), surfaced in sidebar or command palette.
• Better empty states (no results, empty folder, no backlinks).
• Proper 404 page design (currently uses special/error.tpl).

3. JS module boundary (blocks section 4)

Decide before section 4 starts, not during. Section 4 features need non-trivial JS. To avoid proliferating inline <script> blocks that accrete across templates, commit to a module boundary upfront:

• All interactive JS lives in one bundled file loaded once from _emanote-static/, not inlined across templates.
• Templates expose state via data-* attributes; JS reads them. No Heist-spliced JS values.
• Library choice: Alpine.js (declarative, minimal, ~15kb) vs. vanilla Web Components (zero deps, more boilerplate) vs. plain vanilla with a small helper module. Decide as part of scoping section 4.

Existing inline scripts to consolidate under the new module: stork search controller in components/stork/stork-search-head.tpl:17-99, code-copy-button injection in components/styles.tpl:160-198, TOC tracker in components/toc.tpl.

Sequencing

1. Section 1 lands first on top of v4 (visible but orthogonal to behavior).
2. Decide JS module boundary (section 3 scoping).
3. Section 4 features built on the decided boundary.

[srid]

#636