Recover from a 401 by refreshing the token and retrying (no re-login)

[magne4000]

Summary

Makes a mid-session 401 from the Helix API self-heal via the stored refresh token instead of forcing a re-login, and skips a doomed refresh while offline.

Motivation

Today a 401 shows a hard "OAuth token is required" dialog and exits, even when a valid refresh token is still stored — which happens routinely when the short-lived access token lapses (Kodi or the network only just came up after a reboot or an overnight gap). The proactive refresh in Twitch.__init__ doesn't cover a token that expires or is rejected mid-session.

Changes

• api.py error_check (now an instance method): on a 401 with a stored refresh token, force a silent refresh (ensure_valid_token(force=True)) and, on success, raise TokenRefreshed. A one-shot per-instance guard prevents any refresh/retry loop.
• error_handling.api_error_handler (was a passthrough): catches TokenRefreshed, reloads the new token onto the instance, and retries the request once.
• If the retry still 401s while a refresh token survives, it bails quietly (next visit / the proactive refresh recovers). The hard sign-in dialog is now reached only when no refresh token remains — a genuinely dead session.
• Offline guard: while Kodi reports no internet (System.InternetState), ensure_valid_token skips the refresh instead of hanging ~15s on a dead connection; the stored token is kept and refreshed once connectivity returns. A forced post-401 recovery bypasses the guard.
• New TokenRefreshed exception and kodi.connection_available() helper.

Backward compatibility

The private-credential 401 path is unchanged. No settings or stored-token format changes.
The only user-visible change is fewer needless re-login prompts. No new strings.

Testing

• Expire/reject the access token mid-session → list reloads without a re-login prompt
• With no refresh token stored → still prompts sign-in (dead session)
• Offline → no hang; recovers on reconnect
• Normal browsing/playback unaffected