Fix tracking for non-sonnet models

Bugfixes

Improved UI updating lifecycle (frames instead of setInterval)
Fixed some race conditions in the background that could cause it to not load.

Update patchnotes:
- Factors actual style prompt as well when calculating tokens used
- Improved debug logging page (More reliable logging, added option to permanently enable logging)
- Improved/fixed reporting of usage caps

Refactored the entire UI to integrate with the site's UI.

Relevant information is in the chatbox itself, while all information is always in the sidebar.

Fix for android

- Fixed some issues with the MV3 lifecycle (waking up etc).
- Overhauled logging entirely, there is now a fancy debug logs page available from the extension's context menu.

Bugfixes.

Fix issue that was preventing background from waking up properly.

New since the last release:
- Support different tiers (Free/Pro/Team)
- Support multiple accounts via firefox containers (This was hell)
- Support Drive-synced files (Both in projects and chat
- Add option to use the token counting endpoint instead of the gpt_tokenizer estimate
- Now uploads reset times and token amounts to firebase (to crowdsource more accurate counts)

Add a small donation reminder on first install/update.

Minor compatibility improvements

Fixes for incognito mode (disabling it in firefox, using incognito:split in chrome).

Complete rewrite - it now makes API calls instead of inspecting the UI for changes.

This means more types of content are supported (the actual length of user preferences, for one, and any number of project knowledge files as well as the project's custom instructions) and it also supports different features being enabled across different chats automatically, without needing the checkbox menu.

Ensures the sidebar closes after processing the files. Before, the sidebar remained open until the AI was finished sending the response, which meant it would block the view entirely on mobile.