fix(aio): handle socket_action errors in _force_timeout

[disqualifier]

_force_timeout exists to make sure a missing socket or timer event never
stalls indefinitely. It runs in the background and drives
curl_multi_socket_action every ~100ms, so if a wakeup is ever missed, the
next tick recovers it.

Problem

It calls socket_action() without handling errors, and socket_action()
raises CurlError on a non-OK return. Nothing catches it, so the coroutine
ends. It is never restarted, which leaves the safeguard stopped for the rest
of the session. Nothing recovers a missed wakeup after that, so a request
that only needed the next tick to make progress can stall and hang the caller.

Fix

Handle the error inside _force_timeout. If socket_action() fails, report
it as a CurlCffiWarning and keep the loop running, so the safeguard keeps
doing its job instead of ending on the first error. The request-path calls
that raise CurlError on purpose (add_handle and similar) are left
unchanged.

Test

tests/unittest/test_async.py::test_force_timeout_error injects one failing
socket_action call and checks the safeguard is still running afterward. It
fails on main and passes with this change.

Open question

process_data calls self.socket_action() the same way, outside its own
try, so the same CurlError goes unhandled there too. With _force_timeout
fixed its next tick recovers that socket, so I left it alone to keep this
small. Want the same handling on process_data, or a follow-up?

Checklist

• I have manually reviewed the changes and fully understand the code.