Can Kedro Pipeline be stateless?

[duy-rhombus]

Description

Can Kedro support a stateless design where nodes are not recomputed even when a new pipeline is initialized, by leveraging the database model and computation state from a previous run?

Context

In a stateless setup, the pipeline is rebuilt from scratch on every run. This leads to all nodes being recomputed, which is inefficient for costly operations (e.g., LLM calls) and large datasets.

Possible Implementation

Could Kedro allow recovering pipeline computation states (e.g., node outputs) stored externally, such as in S3, to avoid recomputation with a new pipeline instance?