fix: guard optional SparkLLM dependency to prevent pytest collection failure

[mo2g]

Features

• Fix pytest collection failure caused by hard importing SparkLLM when optional dependency sparkai is unavailable.
• Make Spark-related modules degrade gracefully in environments without sparkai, so Full Tests can proceed in CI.
• Keep the existing workflow improvements, but scope this PR mainly to repository code / dependency robustness instead of CI syntax only.

Feature Docs

N/A

Influence

• Affects test collection behavior in CI and local environments without sparkai.
• Reduces coupling between optional LLM provider integrations and the default test path.
• Helps Full Tests run successfully on clean environments where Spark-related dependencies are not installed.

Result

• Before this change, pytest exited during collection with code 4 because SparkLLM was imported unconditionally and sparkai was missing in CI.
• After this change, pytest no longer fails at collection time due to the optional Spark dependency, and Full Tests can continue to actual test execution.

Other

• This PR fixes a repository code / optional dependency handling issue, not just an outdated workflow configuration issue.
• Spark-related functionality still requires installing the corresponding optional dependency in actual usage.