This makes the tests directory an npm workspace and switches to tsup for building our typescript projects.

To test this out:

1. Pull down this branch
2. Run npm install in the root or any workspace (will trigger an npm install for all workspaces)
3. Run npm run dev in js-pkg/react and/or js-pkg/sdk
4. Make some breaking changes
5. Run npm test from tests

Tests should fail.

This standardizes token as the name for a type that provides information for connecting to a resource, comprising of "server tokens" for connecting from the server and "client tokens" for connecting to a particular document from the client.

The fields each token has in common are also standardized.

Previously, we referred to a ClientToken interchangeably as an "AuthDocResponse" and as a "connection key", which is confusing.

It also changes field names to one-word names: doc_id -> doc; base_url and endpoint -> url. This way they are idiomatic on both the rust side and JS side.

This is a breaking change, so I wanted to do it before launching.

Since the fetch handler is exposed externally, anything that imports this as a library will have the fetch handler exposed externally, potentially conflicting with its own fetch handler. This avoids that.

Also splits out the durable object code into its own module, and makes modules public.

We can't easily inject an environment variable into the worker, so a private key for testing is hard-coded in the config.

This PR also makes error handling on the Cloudflare worker side more robust.

Adds a presence hook. Keeps the existing awareness hook to obtain the raw Awareness value (useful for passing to components like CodeMirror), but adds a usePresence hook that works like we discussed.

The demo is just there to confirm that it works; feel free to run with it and turn it into something cooler.

Also adds a github workflow to ensure that prettier has run.

We may also want to run eslint in the workflow, but I'll wait for #3 to land for that as I think it fixes an eslint warning.

We need to return a fully-qualified URL to the client to connect to. The server itself doesn't know by default (or from a request) whether it is behind an encrypting reverse proxy, so it doesn't know whether it should return a ws:// or wss:// URL. The solution is simply to tell the server when we want a wss:// URL instead.

In the native server, this is invoked by passing --use-https when starting the server. For the cloudflare worker, it's invoked by setting the USE_HTTPS variable to true.

This also sets the production config (used for staging) so that it uses wss://.

• Allow region to be specified via env vars.
• Remove some panics and add some error context.
• Fixed an issue where 404s were not being propagated correctly from s3-rust because a feature flag was disabled.

Currently we require a store to be passed on the cli. For local development, it's sometimes nice to use an ephemeral in-memory store without persisting.

This PR changes store from a positional to an explicit argument. If it is not passed, document state is not persisted.

After #56 is merged, which adds a --prod flag, we should probably fail to start with an in-memory store (or at least log an error-level log) if --prod is passed.