feat: add polling-based LLM support

[WH-2099]

Summary

Add SDK support for polling-based LLM invocations.

What Changed

• Added model actions and request models for start_polling and check_polling.
• Added executor routes that dispatch polling calls to LLM implementations.
• Added ModelFeature.POLLING and explicit support checks before polling dispatch.
• Added LLMPollingResult with state-specific payload validation.
• Added tests for request parsing, executor dispatch, unsupported models, and result validation.
• Bumped package version to 0.9.0 and documented polling-based LLM invocation support in README.md.

Validation

• just check
• uv run pytest tests/test_model_polling.py

Closes #332

[gemini-code-assist]

Code Review

This pull request introduces support for polling-based Large Language Model (LLM) invocations. Key changes include the addition of StartPolling and CheckPolling actions to the ModelActions enum, new request and result entities (ModelStartPollingRequest, ModelCheckPollingRequest, and LLMPollingResult), and the implementation of start_llm_polling and check_llm_polling methods within the PluginExecutor. The LargeLanguageModel interface has also been updated with internal methods and a supports_polling check to facilitate this new feature. Feedback provided suggests using value equality (==) instead of identity comparison (is) for StrEnum status checks in the LLMPollingResult validator to ensure more robust logic.