feat: Loop node implementation (#247)

* docs: Add loop node design document

Design for a new loop composite node that iterates until a break
condition is met or max iterations is reached. Includes:
- Do-while execution style
- Parameter flow between iterations
- Lineage tracking via separate branch logs per iteration
- Argo compatibility via recursive templates

Co-Authored-By: Claude Opus 4.5 <[email protected]>

* docs: Update loop node design - branch must always be Pipeline

Clarify that loop node branch parameter must always be a Pipeline,
consistent with other composite nodes (parallel, map, conditional).
Update SDK examples to wrap tasks in Pipeline objects.

Co-Authored-By: Claude Opus 4.5 <[email protected]>

* docs: added plan for loop node

* feat: add LOOP_PLACEHOLDER constant

Add loop placeholder constant following MAP_PLACEHOLDER pattern
for iteration index replacement in loop node names

* feat: refactor placeholder resolution for map and loop

- Rename _resolve_map_placeholders to _resolve_iter_placeholders
- Add support for LOOP_PLACEHOLDER resolution
- Handle nested map and loop placeholders correctly
- Keep old method for backward compatibility

Co-Authored-By: Claude Sonnet 4 <[email protected]>

* feat: add basic LoopNode class with placeholder support

- Create LoopNode class following composite node patterns
- Use LOOP_PLACEHOLDER for branch name generation
- Add validation for break_on and index_as parameters
- Implement _get_iteration_branch_name using refactored placeholder resolution

Co-Authored-By: Claude Sonnet 4 <[email protected]>

* feat: add loop node parameter and branch log management

- Add get_break_condition_value() method with proper validation
- Add _create_iteration_branch_log() using placeholder resolution
- Add _build_iteration_iter_variable() for iteration context
- Proper error handling for missing/invalid parameters
- Add comprehensive tests for all new methods

Co-Authored-By: Claude Sonnet 4 <[email protected]>

* feat: implement LoopNode fan_out method

- Create branch logs for current iteration
- Copy parameters from parent (iteration 0) or previous iteration
- Use proper placeholder resolution for branch names
- Support both initial and subsequent iterations

Co-Authored-By: Claude Sonnet 4 <[email protected]>

* feat: implement LoopNode fan_in method

- Check break condition and max iterations correctly
- Roll back parameters to parent on loop exit
- Set final step status based on branch execution
- Extract current iteration from iter_variable properly

Co-Authored-By: Claude Sonnet 4 <[email protected]>

* feat: loop node, missing argo implementation

---------

Co-authored-by: Claude Opus 4.5 <[email protected]>

Author: vijayvammi

docs/advanced-patterns/async-streaming.md

@@ -1,15 +1,19 @@
-# Async & Streaming Execution
+# 🧪 Async & Streaming Execution (Experimental)
 
-Execute async functions and stream results in real-time with runnable's native async capabilities.
+Execute async functions and stream results in real-time with runnable's experimental async capabilities.
 
-## The Core Insight
+!!! warning "Experimental Feature"
 
-**Runnable natively supports both batch and streaming workflows in a single framework**. While traditional orchestrators like Kedro, Metaflow, and Airflow excel at batch processing, runnable uniquely combines:
+    **Async and streaming capabilities are experimental** and under active development. Features may change, and stability is not guaranteed for production use.
 
-- **Batch Processing**: Traditional data pipelines with full reproducibility and orchestration
-- **Real-Time Streaming**: Native AsyncGenerator support for LLM inference, APIs, and streaming data
+## Overview
 
-**AsyncPipeline and AsyncPythonTask** bring this streaming capability natively to runnable, enabling real-time workflows while maintaining all of runnable's core features: parameter management, catalog system, reproducibility, and multi-environment execution.
+**Runnable provides experimental async and streaming support** for specialized use cases requiring real-time processing:
+
+- **Batch Processing**: Core production-ready data pipelines with full reproducibility and orchestration
+- **Experimental Streaming**: AsyncGenerator support for LLM inference, APIs, and streaming data
+
+**AsyncPipeline and AsyncPythonTask** are experimental features that enable streaming workflows while maintaining runnable's core features: parameter management, catalog system, reproducibility, and multi-environment execution.
 
 !!! warning "Local Execution Only"