release: Upgrade version to v1.3.37

[WCY-dt]

This release adds configurable message delivery, reversible advertising filtering, and a more usable visual XPath picker.

Notifications have their own settings tab with bundled Telegram, Discord and Feishu/Lark logos, multiple destinations, shared quiet hours/timezone settings, independent rule editors, connection tests, previews and delivery history. New-article alerts and scheduled briefings support inclusion/exclusion conditions, recipient selection, article limits, weekdays, delivery times, cached summaries or feed excerpts, and source links. Encrypted credentials, atomic queue/cursor persistence, bounded retries, destination-change cancellation and existing proxy settings work in desktop and server modes. Notification delivery is opt-in.

Advertising controls live in Content settings and reuse existing AI profiles, provider protocols, credentials, proxies and shared usage limits. Basic filtering handles dedicated ad resources and explicit network markers; ambiguous text goes through opt-in AI review. The model selects bounded article blocks with exact evidence, a second confirming audit and broad-removal safeguards. Default review mode waits for Apply; automatic mode is optional. Both modes preserve the original content, support site exceptions, cancellation, cached decisions and one-click restoration. Advanced domain-scoped CSS rules and local/AI sample previews are available. AI applies to RSS and extracted full text; the webpage proxy supports basic filtering only, and external browsers are unaffected. No extensions are installed. Model confidence and two-pass agreement are not presented as measured accuracy guarantees.

The XPath picker now supports undo/redo, simplified layouts, parent/child navigation and keyboard confirmation, paginated extraction results, missing-field diagnostics and locating a result in the preview. Consistent unique dates/images are suggested automatically. Applying a selection preserves manual author/category expressions and custom IDs. Settings search, English/Chinese translations, draft protection, light/dark themes and narrow-screen layouts cover the new controls.

Validation:

• All backend packages passed go test -timeout=5m ./internal/...; affected packages were rerun after final changes. go vet ./... passed.
• Race detection passed for advertising filtering, core/settings/media handlers; earlier notification/database race checks also passed.
• Full frontend suite passed 236 tests, followed by an 8-test targeted run including the additional empty-filtered-full-text restoration regression. ESLint reported no errors.
• Compatible Wails v3 desktop and server builds passed on Windows.
• Computer/browser inspection used an isolated database and local mock model: settings saving/draft protection, review/evidence/apply/restore, XPath inference/pagination/locating/undo/redo, light/dark themes and 390px/1280px layouts. Runtime inspection caught and fixed a translation parser error and a lost XPath undo candidate.
• Tests cover configuration round trips/conflicts, original-content preservation, sanitization, model response validation, shared AI accounting/limits, cancellation/stale responses, site exceptions and resource blocking before network access. Deterministic mocks verify behavior; no live-model accuracy claim or real-provider delivery claim is made.

Usage and scope are documented in docs/NOTIFICATIONS.md, docs/AD_FILTERING.md and both XPath guides. Server OpenAPI and the generated skill API reference include the new endpoints. The pre-existing local package-lock changes are excluded.