Release v2.8.0

🚀 New Features

OpenAI Codex search backend — 19th search provider. Uses Pi-managed authentication via AuthStorage.getApiKey (no API key in config). Run /login in Pi and select OpenAI Codex. Supports configurable model (default: gpt-5.4-mini). Hosted web_search injected via injectCodexSearchPayload.
Targeted combine mode — New combineMode: "targeted" option caps combine fan-out to 3 usable backends instead of querying all enabled backends. Uses configured selection strategy to order backends, queries in batches until 3 return non-empty results, then RRF-merges. Set in search.json: { "combine": true, "combineMode": "targeted" }.

timeoutSignal wrapper added to OpenAI Codex stream call — only backend missing the 30s timeout guard.
combineMode validated at runtime — unrecognized values now log a warning and fall back to "all" instead of silently falling through.
search-status display fixed for Pi-managed auth backends — shows "— (Pi auth)" instead of a misleading "key: ✓" checkmark.
Content field used as fallback in normalizeSearchResult when snippet is missing — valid results no longer silently dropped.

164 tests passing (13 new: URL helpers, content fallback, targeted combine edge cases).
URL normalization helpers exported and unit-tested — normalizeHttpUrl, normalizeUrlForDedup, looksLikeDomainOrPath now directly tested.
runTargetedCombine edge cases — empty backends, all fail, single backend, numResults distribution.
Content fallback test — verifies results with content but no snippet are preserved.