Performance: Cache node instances in Runtime worker to avoid repeated instantiation

[coderabbitai]

Issue Description

Currently in the Runtime._worker method, a new node instance is created for each execution:

node = self._node_mapping[state["node_name"]]
outputs = await node()._execute(node.Inputs(**state["inputs"]))

This approach ensures isolation but may impact performance, especially under high throughput scenarios.

Proposed Solution

Consider caching node instances to improve performance:

• Cache instances in a dictionary keyed by node name
• Reuse existing instances if nodes are stateless
• Manage a pool of instances if needed

Implementation Considerations

• Evaluate if nodes are stateless and safe to reuse
• Consider thread/async safety implications
• Measure performance impact before and after changes

References

• Identified in PR Cleaning exospherehost python sdk and adding first exospherehost runtime #141: Cleaning exospherehost python sdk and adding first exospherehost runtime #141
• Original comment: Cleaning exospherehost python sdk and adding first exospherehost runtime #141 (comment)
• File: python-sdk/exospherehost/runtime.py around lines 289-290