Cherry-picked from PR #3695 by binhnt92.
Matrix _sync_loop() and Mattermost _ws_loop() were retrying all errors
forever, including permanent auth failures (expired tokens, revoked
access). Now detects M_UNKNOWN_TOKEN, M_FORBIDDEN, 401/403 and stops
instead of spinning. Includes 216 lines of tests.
save_config_value() used bare open(path, 'w') + yaml.dump() which truncates
the file to zero bytes on open. If the process is interrupted mid-write,
config.yaml is left empty. Replace with atomic_yaml_write() (temp file +
fsync + os.replace), matching the gateway config write path.
Co-authored-by: Hermes Agent <hermes@nousresearch.com>
PR #3323 added retry with exponential backoff to cache_image_from_url
but missed the sibling function cache_audio_from_url 18 lines below in
the same file. A single transient 429/5xx/timeout loses voice messages
while image downloads now survive them.
Apply the same retry pattern: 3 attempts with 1.5s exponential backoff,
immediate raise on non-retryable 4xx.
binhnt92