fix(security): prevent XSS in library search snippets

[farce1]

Motivation

• The library search returned SQLite FTS snippet(...) output containing raw <mark> HTML that was rendered via dangerouslySetInnerHTML, allowing attacker-controlled meeting titles/transcripts to inject executable HTML/JS.
• The goal was to remove unsafe HTML injection while preserving visible search highlight behavior and minimal surface change to existing UI.

Description

• Added non-HTML sentinel constants SEARCH_SNIPPET_HIGHLIGHT_START and SEARCH_SNIPPET_HIGHLIGHT_END in src/lib/searchSnippet.ts to be used as FTS highlight markers instead of <mark> tags.
• Updated the FTS query in src/hooks/useLibrary.ts to request snippets with the sentinel markers by switching the snippet(...) call to snippet(meetings_fts, 1, $2, $3, '...', 32) AS snippet and passing the new sentinel constants as query parameters.
• Replaced unsafe dangerouslySetInnerHTML usage by removing renderSearchSnippet from src/views/LibraryView.tsx and changing SearchResultRow (src/components/library/SearchResultRow.tsx) to a safe renderer that splits the snippet on the sentinel tokens and constructs text nodes and explicit <mark> React nodes, ensuring any HTML in content is treated as text.
• Kept the highlighting and visual behavior intact while eliminating injection of raw HTML into the DOM.

Testing

• Ran a production build with npm run build, which completed successfully.
• Executed unit tests for the touched library logic with npm run test -- src/hooks/useLibrary.test.ts, and all tests passed (1 file, 5 tests).

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0c2fc83db8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Handle literal highlight sentinel collisions

renderSnippet parses any occurrence of SEARCH_SNIPPET_HIGHLIGHT_START/END as markup boundaries, but those tokens now come from normal searchable content too, so a meeting transcript/title that literally contains these strings is mis-rendered. With an unmatched start token, the marker is stripped and text can be dropped (end === -1 path), and with both tokens present, arbitrary content is shown as highlighted even when it was not matched, which makes search results misleading for user-controlled data.

Useful? React with 👍 / 👎.