[secretpray]

Summary

• Safe failover and retries. A request that changes something (starting or stopping a session, setting prices) is resent, to the same or another API host, only when the connection could not be established. Before, a timeout sent it to the next host, which could start a session twice.
• Retries after network errors. GET, PUT and DELETE requests are retried after connection errors and timeouts, not only after error responses.
• Retry-After. The wait is capped by the new max_retry_after option (OCTA_MAX_RETRY_AFTER, 60 seconds by default); a longer wait raises RateLimitError at once. The header is also accepted as an HTTP date.
• Errors. HTTP 400 raises ValidationError (a subclass of APIError, so existing except APIError keeps working); HTTP 409 raises the new ConflictError. Every transport failure raises APIConnectionError or APITimeoutError instead of an httpx exception.
• Configuration. ConfigurationError is also a ValueError; invalid OCTA_* numbers raise it with the variable name. The API token no longer appears in repr(ClientConfig).
• Guides, docstrings and the changelog describe the new behaviour.

Verification

• ruff check, ruff format --check, mypy (strict), pytest: 78 passed, 100% coverage
• new tests for each case: POST after a timeout and after a connection failure, GET retry after a connection error, Retry-After cap and HTTP date, 400/409 mapping, token not in repr, invalid environment numbers
• every code example in the README and the guides runs against the package with mocked HTTP