fix: replace print() with logger and strengthen error handling

[crazywriter1]

Summary

• alpha.py — Replace two print() calls with logger.warning() so gas estimation failures and scheduler registration errors are captured by the caller's logging infrastructure instead of going to stdout
• alpha.py — Change raise Exception(...) to raise RuntimeError(...) in new_workflow to match the exception type used consistently everywhere else in the codebase; updated docstring to reflect that gas estimation failure uses a fallback instead of raising
• alpha.py — Fixed _register_with_scheduler docstring: previously claimed Raises: Exception but the method never raises — errors are caught, logged, and swallowed intentionally since the workflow contract is already deployed
• llm.py — Strengthen the choices[] guard in _chat_request: added isinstance(choices, list) and isinstance(choices[0], dict) checks — the previous if not choices guard passes silently for [None] or non-list types, which then crashes with an unhelpful AttributeError
• llm.py / types.py — Log malformed SSE JSON chunks with logger.warning() instead of silently continue-ing, making it visible when the server sends broken data during streaming

Files Changed

• src/opengradient/client/alpha.py
• src/opengradient/client/llm.py
• src/opengradient/types.py
• tests/client_test.py

Test Plan

• Full test suite passes: uv run pytest tests/ -v → 121 passed
• TestChoicesGuard — verifies choices=[] and choices=[None] both raise RuntimeError
• TestSSEJsonLogging — verifies malformed SSE JSON emits logger.warning, valid chunks do not
• TestAlphaErrorHandling — verifies deployment failure raises RuntimeError, gas estimation failure and scheduler failure both call logger.warning

[adambalogh]

Hey, would you mind merging from main? Thanks!

[crazywriter1]

Done. Merged from main and force pushed. Thanks for the review. @adambalogh