fix(api): improve connection resilience and handle connection resets

[tst-038]

Summary

Resolves unhandled requests.exceptions.ConnectionError: ('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer')) when communicating with Thermia Online cloud during register writes or transient connection drops.

Details

• Added a 15-second timeout to all HTTP requests.
• Added a _request_with_retry wrapper that catches socket disconnects (ConnectionResetError, ConnectionError, Timeout).
• When a connection drops, the broken session is closed and a fresh session is created before retrying the call.
• HTTP status errors (like 400 or 500) are not retried—only actual transport/socket disconnects trigger a retry.
• Added a unit test (test_retry_resilience.py) simulating a socket reset during a POST to ensure it recovers and succeeds.

[janfryauf]

Thanks for working on this. I ran into the same hang (klejejs/ha-thermia-heat-pump-integration#147) and opened #63 with only the timeout part: a default timeout on the session and fewer retries, about 20 lines.

I kept it separate because this PR also retries requests that aren't safe to repeat (the SelfAsserted login POST, the single-use auth-code exchange and register writes, so a write to the heat pump could be sent twice), and it changes __set_register_value to raise on non-200 responses. Both PRs edit the same lines, so only one can go in as is. Happy to close mine if you'd rather narrow this one down to the timeout.

[tst-038]

Thanks for the feedback! Looking at your PR #63, using ThermiaSession to set the default timeout and letting urllib3.Retry handle safe GET retries is definitely the cleaner and more idiomatic approach.

Let's go ahead with your PR #63 for the timeout fix.

I'll close this PR and open a small, separate PR specifically for __set_register_value raising on failure so it can cleanly pair with ha-thermia-heat-pump-integration#146
without mixing concerns. Thanks again!