Enhance search functionality with regex, FTS5, snippets, and sorting

[jesserobbins]

Fix and Improve search accuracy and performance across full-text, fast (Parquet), and aggregate query paths with word-boundary matching, better snippet handling, and consistent result sorting.

Problem

While importing my old 1990's era email archives, I discovered a number of issues with search

1. Substring matching causes false positives — Text search matches partial strings within words, returning irrelevant results
2. Body text invisible in fast search — Messages where search terms only appear in the body don't show in TUI unless user switches to deep search mode (which I feel is underdocumented but that's a separate issue)
3. Inconsistent sort behavior — Search results don't respect the user's selected sort field (Name/Count/Size)

Proposed Solution

• Word-boundary regex (\b) for text search to match exact words, not substrings
• Prefix matching in FTS5 (* operator) for partial word matches within the deep search path
• Snippet inclusion in fast search conditions so body text matches appear immediately in TUI

Changes

• internal/query/sqlite.go — FTS5 prefix matching
• internal/query/duckdb.go — Word-boundary regex,
  snippet inclusion, sort parameter, escapeRegex helper

Testing

• Text search with word boundaries prevents false
  substring matches
• Parquet fast search now catches body text matches

[roborev-ci]

roborev: Combined Review (945c419)

Summary verdict: Changes are not ready to merge due to a compile-breaking API migration and an ignored sorting parameter.

High

• Location: internal/query/engine.go:31
• Problem: Search, SearchFast, and SearchFastWithStats now require MessageSorting, but the change is not propagated to existing call sites and implementations, including internal/mcp/handlers.go:131, internal/api/handlers.go:1196, internal/tui/model.go:498, internal/remote/engine.go:575, and internal/query/querytest/mock_engine.go:99. The package will not compile.
• Fix: Update all callers, the remote engine, mocks, and tests to pass or accept the new MessageSorting argument, or keep the old interface until the full call chain is migrated.

Medium

• Location: internal/query/sqlite.go:1207
• Problem: The new sorting argument is accepted but never used. Search results still hard-code ORDER BY m.sent_at DESC, and the DuckDB search/cache paths likewise keep fixed date-desc ordering. Callers passing size, subject, or ascending sort options will silently get the wrong order.
• Fix: Thread sorting into the query builders and page cache helpers, generate a validated ORDER BY consistent with ListMessages, and include sorting in the DuckDB search cache key if cached result order can vary.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[jesserobbins]

Removed the incomplete sort stuff

[roborev-ci]

roborev: Combined Review (d9e2366)

High-risk search regressions found: DuckDB regex boundaries miss punctuation-leading terms, and SQLite FTS5 can misparse unquoted special characters.

High

• Location: internal/query/sqlite.go:1171-1175
  Problem: FTS5 search terms are only double-quoted when they contain a space. Terms containing FTS5 special characters such as -, :, (, or ) can be interpreted as query syntax instead of literals, causing syntax errors or incorrect results.
  Fix: Quote all FTS terms before appending the prefix operator, e.g. ftsTerms[i] = fmt.Sprintf("\"%s\"*", term), and remove the conditional strings.Contains(term, " ") branch.

• Location: internal/query/duckdb.go:461-462, internal/query/duckdb.go:2393-2394
  Problem: Prefixing escaped search terms with \b causes DuckDB regex searches to miss terms that start with non-word characters, such as +15551234567, @gmail.com, #bug, or punctuation-prefixed tokens. Because \b requires a word/non-word transition, it fails when both the preceding character and the first character of the term are non-word characters. This regresses cases that previously matched via ILIKE.
  Fix: Only apply the word-boundary prefix when the term starts with a word character, or fall back to substring matching / a boundary expression that handles start-of-string and non-word separators for punctuation-leading terms.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (e31a46d)

Verdict: No Medium, High, or Critical issues were reported.

All agents that provided findings agree the diff is clean; no actionable findings to include.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)