A stale config can hold a refresh token issued by a Tidal client that
no longer exists ("Client id ... not found") next to an access token
that is still perfectly valid. The proactive refresh in
ensure_fresh_token treated its own failure as fatal, so every request
died on the unusable refresh token without ever trying the working
access token.
Classify refresh failures: a 4xx from the token endpoint is permanent,
so warn once, stop proactive refreshing, and keep serving with the
current access token; transport errors and 5xx are transient and keep
the retry metadata. The 401-retry in make_request stays as the backstop
for an access token that has actually expired, and a restart heals the
state via the device-login fallback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
9706e3b5d2