More reliable REST, MCP and gRPC access

Calls arriving over the REST API or MCP without a resolvable session — an anonymous request, or a client that sends no session header — were handled on the mesh's internal routing layer instead of a real portal session. Those calls could hang until they timed out rather than returning an answer. Every API and MCP call now runs on a proper portal session, and the same applies to the GitHub sync operations started from MCP.

For gRPC, validating an API token now happens on a portal session too. Previously a failure in that step was quietly treated as "not signed in", so a valid token could silently lose its identity and the caller would see permission errors instead of a clear failure.

The Python and TypeScript SDKs now reuse one connection identity for the lifetime of the process instead of generating a new one on every connect. Reconnecting keeps the subscriptions and live streams you already set up, rather than leaving them behind on an abandoned session. Set MESHWEAVER_CLIENT_ID if you want that identity to stay the same across restarts as well.

Reconnecting…
The connection to the server was interrupted. Trying to restore it…
Trying again…
The connection could not be restored. Reloading the page…
The server was updated. Reloading the page to pick up the latest version.