Improve checkpoint versioning stability

[yl-lisen]

Use case

In proton, query state management is works via checkpoints. currently we use a strict checkpoint/recover implementation, which is easily broken when there are any modifications about query pipeline, processors or its header changes.

So we want to improve the stability of checkpoints, and reduce broken scenarios as much as possible

Describe the solution you'd like

Describe alternatives you've considered

Additional context

[jovezhong]

(Jove Github Bot) added it to the current sprint.