feat: Report the auto-configuration stream's health in the status resource

[keelerm84]

Summary

A back-port of v9's auto-configuration status reporting, so the two lines report the same thing the same way.

v8's status resource says nothing about the auto-configuration stream. A Relay Proxy whose auto-configuration key has been rejected therefore reports itself healthy: every environment it already knows about is fine, per-environment connectionStatus looks normal, and the stream's own state appears nowhere in the document. The only signal an operator has is a log line.

This adds autoConfigStatus, present only in automatic configuration mode:

"autoConfigStatus": {
  "state": "INTERRUPTED",
  "stateSince": 1757000000000,
  "lastError": { "kind": "ERROR_RESPONSE", "statusCode": 503, "time": 1757000000000 }
}

It reuses interfaces.DataSourceState and ConnectionErrorRep, matching v9, so the stream reports the same states and error kinds as the per-environment connectionStatus and a reader only has to learn one vocabulary.

ConnectionErrorRep gains the HTTP status code

v9 already has this field; v8 did not. Without it a failing stream shows up as a bare ERROR_RESPONSE with no way to tell a 401 from a 503, which is exactly the distinction an operator needs. It is populated for the per-environment error as well, not only the new one, so the field is not half-filled.

Two behaviors worth reviewing

An interruption while the stream has never connected keeps INITIALIZING. This matches the SDK data source status: a first connection that never succeeded must not report that it was once working. The error is still recorded, so the reason is visible.

Reporting VALID requires a handled event, not a connection. Dispatching an event creates environments, starts SDK clients and writes the cache, and the connection can die while that runs — so an event is not by itself evidence the stream still works. The failure count is read before the dispatch and success recorded only if nothing failed in the meantime. This is v9's failureGeneration/markValid mechanism; I would not have arrived at it independently.

Scope

Stream behavior is unchanged. A rejected key still reports OFF, because on this line the stream stops and the process exits.

The auto-configuration retry work (#866) is stacked on top of this and changes that one state to an interruption, since the stream then keeps trying instead of giving up. That PR had grown its own ad-hoc version of this reporting; it has been dropped in favour of v9's shape, and splitting it out here keeps both diffs to one idea each.

---

Note

Overview
Adds autoConfigStatus to the Relay status JSON when running in automatic configuration mode, using the same DataSourceState / connectionStatus shape as per-environment streams so operators can see when Relay has stopped learning new environment config (e.g. rejected key or stream errors) even if existing envs still serve flags.

StreamManager now tracks connection state (initializing, valid, interrupted, off), records failures with HTTP status codes where applicable, and only moves to VALID after an SSE event is fully handled—via a failure-generation check so a disconnect during dispatch cannot falsely report a healthy stream. Interruptions before the first successful event stay INITIALIZING, matching SDK data-source semantics.

ConnectionErrorRep gains optional statusCode, populated for both the new auto-config block and existing per-environment lastError fields.

Unit tests cover valid-after-event, HTTP 503/401 handling, and terminal OFF on close or bad key.

^{Reviewed by Cursor Bugbot for commit 69aa10c. Bugbot is set up for automated code reviews on this repo. Configure here.}