Switch permissions fetching logic to use a snapshot, not getDoc.

The logic was finicky when firestore thinks it's disconnected (e.g. large cards in production on a slow connection that take more than 10 seconds to load). We added retry logic to keep trying to fetch permissions, but that was finicky and also sometimes didn't fire.

The better appraoch is to use onSnapshot, which will fire when the connection is live automatically.

Part of #690.

Coded by Claude.

Fix reindexCardEmbeddings bug.

It didn't handle the version bump use case correctly.

Claude.

Part of #723.

Update to use the most modern / capable models.

Make external links in Bulk Export be included in output.

Part of #723.

Vibecoded by Claude.

Failed messages get a retry button.

Part of #710.

Stream responses in chats back to the client.

Part of #710.

Merge branch 'chat-streaming-sse'

* chat-streaming-sse:
  Actually show the streaming message in the client.
  Actually start streaming tokens to the client.
  Refactor it so the assistant message is not kicked off by createChat or postMessage.
  postMessageInChat checks to verify that the last message in the thread is an assistant message.
  Switch createChat to also use the new OnRequest structure.
  Switch to using onRequest for postMessageInChat.
  Create an authFromRequest helper function.
  Make throwIfUserMayNotUseAI take a uid.

mulit-edit dialog allows bulk publishing / unpublishing.

Fixes #717.

Tooltip on chat title.

To make it possible to read the whole title if it's too small.

Part of #710.

Merge branch 'chat'

Part of #710.

* chat: (149 commits)
  Fix it so anonymous people can view published chats.
  Add published view and toggle in the UI.
  Add a togglePublishedForCurrentChat action craetor.
  Add a VISIBILITY_OFF_ICON.
  Update firestore rules around published chats and messages.
  Add a published field to chat, defaulting to false.
  Automatically set a title for the chat by asking AI.
  Bring back the styling of the user/asssitant labels claude code originally gave.
  A visual overhaul for chat courtesy of Claude Code.
  The initial chat dialog shows up as a nice compose dialog.
  Render the list of cards in the chat as a zippy.
  Store the prompt version in the chat message.
  Filter out cards the user can't view in createChat.
  Remove an unnecessary points.sort.
  Remove gpt-tok from functions.
  createChat sorts by similarity of the initial message to cards.
  updated timestamp of chat is updated when a new message is posted.
  Up the memory limit for createCard so it doesn't OOM.
  Fix the import.
  createChat now fits as many cards as fit in the amount of context window we were provided.
  ...

Merge branch 'claude'

Part of #710.

* claude:
  Update selectUserMayUseAI to report true if openai or anthropic is enabled.
  Switch to use anthropic model by default.
  Implement claude client endpoint.
  Add in a token count esitmator for claude.
  Wire in stubs for where anthropic will connect to the anthropic function.
  Install latest zod to fix build errors.
  Clean up the FIREABSE_DPELOY_TASK to be easier to grok.
  Create the anthropic cloud function endpoint.
  Allow a user passing an Anthropic API key.