Allow to customize reasoning

Connects directly to the Codex App Server without requiring ChatGPT account metadata.
Supports custom model selection in Settings, including OpenRouter models.
Preserves capability-token authentication over the configured WebSocket.
Updated Firefox and Chrome extension packages.

Fix to force stop and start over

Changes since 1.20.1:

• Improved performance and reduced memory usage during large discovery and drafting sessions.
• Moved drafts, source posts, reply history, and profile revisions to IndexedDB.
• Added safe, resumable migration of existing data. Existing profiles, drafts, edits, and reply history are preserved.
• Added stable draft pagination with first, previous, direct page, next, and last-page navigation.
• New drafts no longer unexpectedly change the page currently being reviewed.
• Added a notification that lets users merge newly received drafts when ready.
• Remembered the current page and scroll position for each profile filter.
• Draft edits now survive paging and extension restarts.
• Added protection against silently overwriting drafts edited in another panel.
• Queue pages now load independently of the Codex connection and usage-status checks.
• Reduced repeated storage writes and limited loaded queue data to the current and adjacent pages.
• Improved X page scanning by processing new or changed posts instead of repeatedly scanning every visible post.
• Improved discovery deduplication, progress tracking, and candidate batching.
• Preserved failed and retryable draft records instead of automatically reducing older records.
• Improved handling of long-running discovery sessions and large draft histories.
• Added matching Firefox and Chrome behavior, automated parity checks, expanded tests, and performance benchmarks.
• Fixed several pagination, scrolling, stale-response, and queue-update race conditions.

Publishing, account switching, liking, and following remain manual. No automatic X actions were added.

• Added profile creation wizard, copying, deletion, and full profile editing, including X account URLs and custom AI context.
• Improved Welcome with clear setup instructions, profile creation, and easy navigation back after creating a profile.
• Expanded single-profile discovery toward a configurable target of 100 ready drafts, with deeper searches and related terms grouped using OR.
• Highlighted posts found while browsing or scrolling X can now contribute to the active discovery session.
• Added profile filtering and pagination for drafts, with 25 items per page and consistent navigation buttons.
• Improved draft layout, long-URL wrapping, and preservation of unsaved edits as new drafts arrive.
• Added Arctic Daylight as the default light theme and Ocean Signal as an optional dark theme.
• Unified confirmation dialogs across profile management and AI replies.
• Added an About page with project descriptions, logos, websites, and author profiles.
• Simplified Codex setup with Windows, macOS, and Linux instructions. Removed the legacy server component; Xboost connects directly to Codex App Server.
• Fixed profile creation from Welcome, hidden-button visibility, and several layout issues.

Reply publishing, likes, follows, and X account switching remain manual.