fix(voice): keep a stopped round's actions and follow-up memory

A reply stopped partway through a round now keeps the actions that ran,
answering the calls it never reached as skipped, so the history stays
paired for both providers. A failed typed turn no longer resets the
pointer, referents and follow-up memory, since its conversation continues.

The server now tells a rejected key (401, or Gemini's 400 with reason
API_KEY_INVALID, matched exactly), denied access (403) and an account out
of credit (OpenAI's 429 insufficient_quota) apart; each cools down for a
minute and gets its own message in the chat.

docs/CURRENT-STATE.md

@@ -703,9 +703,13 @@ The chat routes answer a missing key with `503` and code `API_KEY_MISSING`.
 OpenAI chat can only bill `OPENAI_API_KEY`, so while `GEV_PREFER_CODEX_OAUTH`
 pins OpenAI to ChatGPT OAuth it answers `503` with code `CODEX_OAUTH_REQUIRED`
 instead; Gemini chat is unaffected. An upstream `429` or `503` keeps its status
-and passes on a `Retry-After` given in seconds, capped at an hour. After an
-upstream `401` or `403`, the route answers `502` without calling the provider
-for a minute, unless the configured key changes.
+and passes on a `Retry-After` given in seconds, capped at an hour; the chat
+names the wait, and a throttled reply carries code `RATE_LIMITED`. A refused
+key or account stops reaching the provider for a minute, unless the configured
+key changes: `API_KEY_REJECTED` for a `401` or Gemini's `400` with reason
+`API_KEY_INVALID`, `ACCESS_DENIED` for a `403`, and `QUOTA_EXCEEDED` for
+OpenAI's `429` `insufficient_quota`. Each refusal answers `502` with its code,
+and the chat shows a matching message.
 
 Chat and the voice panel sit above the dock or dock into the right rail's lane,
 which hides the rail until the panel minimizes or closes, or a rail panel opens.