fix(api): defer streaming response until referenced variables are updated

[tomerqodo]

Benchmark PR from agentic-review-benchmarks#1

[augmentcode]

🤖 Augment PR Summary

Summary: This PR fixes streaming responses that reference conversation variables by ensuring output streaming is deferred until the relevant variable assignment has completed.

Changes:

• Added blocks_variable_output() to the v1 Variable Assigner node so the streaming coordinator can treat it as a blocking node for specific variable selectors.
• Adjusted v1 Variable Assigner NodeRunResult.inputs["value"] handling to return a value appropriate to the write mode (notably for clear).
• Added a new workflow fixture (test_streaming_conversation_variables_v1_overwrite.yml) that assigns sys.query into a conversation variable and then renders it in an Answer node.
• Added a unit test validating that streamed conversation-variable chunks reflect the updated value (and not the default) when using v1 overwrite assignment.

Technical Notes: The response coordinator builds a set[tuple[str, ...]] of referenced variable selectors from response templates, and uses blocks_variable_output() to decide which upstream nodes must finish before flushing streaming output.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 1 suggestions posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

variable_selectors is built as a set[tuple[str, ...]] (see response coordinator), but assigned_variable_selector is a Sequence[str] (likely a list), so assigned_selector in variable_selectors will never match and this node won’t actually block streaming. Consider normalizing to a tuple (similar to v2.VariableAssignerNode.blocks_variable_output) before the membership check.

_{🤖 Was this useful? React with 👍 or 👎}