feat: search body-text indexing with snippet extraction

[Shewart]

Summary

Extends the search index from title/description-only to full body-text matching, with a 2-line snippet surfaced when the match wouldn't otherwise be visible. Same client-side substring scoring, no build-step change, no new dependencies.

What's new

MarkdownPlainText.Extract(markdown, maxLength = 8000) in ShellDocs.Core

Strips markdown/HTML noise down to searchable prose:

• YAML frontmatter (leading ---…---)
• Fenced code blocks (whole block)
• HTML tags and Razor component tags (<Callout Title="x">body</Callout> → body)
• Images ![alt](/img) → alt
• Links [text](url) → text
• Inline code (backticks removed, content preserved)
• Emphasis (**bold**, *italic* → bold, italic)
• Heading # prefixes (heading text kept)
• List markers (-, *, 1.) and blockquote >
• Whitespace collapsed, trimmed to maxLength

SearchEntry.Body (nullable) — populated for Page entries during SearchIndex.FromGraph. Heading entries stay body-less (they anchor into the same page).

SearchDialog scoring extended

Body substring match adds +6; per-token body match adds +1. Both weighted lower than title/description/section so page name still wins on ties. Hard-reject still applies — every token must appear somewhere.

Body-only match snippet

When the match comes from body (not title/section), the result surfaces a 2-line snippet centered on the first token hit:

Search: "static site"
Result: Deploying
        ShellDocs
        … output as a **static site** ready for GitHub Pages, Vercel, Netlify …

Snippet is 150 chars, ellipsis prefix/suffix when trimmed, case-preserving. .search-result-snippet CSS: 0.75rem muted, -webkit-line-clamp: 2 for the two-line box.

Out of scope (deliberate)

• Highlighting inside snippet — no <mark> around the matched substring yet. Polish; its own follow-up.
• Per-heading body splits — body is per-page today. A match near the bottom of a long page anchors the URL at the top. Splitting body at heading boundaries and emitting per-section entries is a real improvement, but doubles the index size and complicates the anchor URLs.
• Real inverted index / BM25 — substring scoring is fine for docs-sized corpora (~100 pages, seen in shelldocs-preview). If someone builds a 1000-page site with this, we'll rebuild the ranker.

Test plan

• dotnet build shelldocs.slnx — clean, 0 warnings, 0 errors
• dotnet test shelldocs.slnx — 139 / 139 passing (+10 new: 8 MarkdownPlainTextTests, 2 SearchIndexTests)
• Preview app: all primitive pages + /docs/introduction render clean — SearchEntry schema change (added optional Body field with default null) didn't regress anything downstream
• Manual: Cmd/Ctrl+K, search for a term that appears only in a page body (not any title or section) — result surfaces with a 2-line snippet under the title, URL navigates to the correct page

New test coverage

MarkdownPlainTextTests (8):

• Strips frontmatter
• Strips fenced code blocks (preserves surrounding prose)
• Strips Razor component tags (preserves body)
• Unwraps links and images (keeps text, drops URL)
• Unwraps emphasis and inline code
• Strips heading # prefixes (keeps heading text)
• Truncates at maxLength
• Empty/null input safe

SearchIndexTests (+2):

• Page entries carry extracted body text
• Heading entries have null body (dedup — they anchor into the page's body)

Files touched

• src/ShellDocs.Core/MarkdownPlainText.cs (new)
• src/ShellDocs.Core/SearchIndex.cs — SearchEntry.Body field, ExtractBodyFromFile in Walk
• src/ShellDocs.Components/Chrome/SearchDialog.razor — body scoring, snippet extraction
• src/ShellDocs.Components/Chrome/SearchDialog.razor.css — .search-result-snippet styling
• tests/ShellDocs.Tests/MarkdownPlainTextTests.cs (new)
• tests/ShellDocs.Tests/SearchIndexTests.cs (+2 tests)