Make git_cacher resilient to connection timeouts

[anfimovdm]

GiteaClient.make_request only caught ClientConnectorError and ServerDisconnectedError, so a ConnectionTimeoutError (or any other ClientConnectionError subclass) propagated up, cancelled the asyncio.gather in git_cacher.run, killed the process and triggered a container restart loop.

• Broaden the retry catch to aiohttp.ClientConnectionError and asyncio.TimeoutError so timeouts are retried like other transient network errors.
• Return [] explicitly after exhausting retries; the function used to fall off the end and return None, which crashed the caller with TypeError on items.extend(None).
• Use return_exceptions=True on the gather calls in git_cacher so a single bad repo or organization is skipped instead of taking the whole cacher down.