[Enhancement] Refactor graph configuration management and custom config support

[zephyrzilla]

Summary

This PR fixes multiple configuration management issues in the graph execution pipeline. It addresses problems with redundant config reloads, improves override handling, and adds support for custom model configurations. The changes introduce AutoNestedDict for better nested config manipulation and refactor the config flow to pass dictionaries directly instead of file paths, eliminating unnecessary I/O operations.

Features:

• Graph configuration loading: Now supports direct config dict passing and existing task detection
• Model configuration: Custom model configs can override default models.yaml values
• Configuration overrides: .override() method now applies changes directly to config dict
• Graph properties: Cached in memory instead of reloaded from disk on every node access
• Workflow execution: Both programmatic and YAML-based tasks follow unified config handling

How to Test

Test 1: Existing YAML Task Execution

1. Navigate to an existing task directory with graph_config.yaml
2. Run: Workflow(name="<task>").run(num_records=5)

Test 2: Programmatic Workflow with Overrides

1. Create a workflow programmatically:

   from sygra import Workflow
   from sygra.utils.utils import load_model_config

   cfg = load_model_config()
   model_cfg = cfg.get("gpt-4o")
   model_cfg["name"] = "gpt-4o"

   wf = Workflow("test_task")
   wf.source({"type": "hf", "repo_id": "glaiveai/glaive-code-assistant-v2"})
   wf.llm(model="gpt4", prompt="Tell me a fact: ")
   wf.override("graph_config.nodes.llm_1.model", model_cfg)
   wf.run(num_records=1)

2. Verify the override is applied

Checklist

• Lint fixes and unit testing done
• End to end task testing
• Documentation updated

[psriramsnc]

LGTM 🚀