fix: reusable-vm runs dispose interceptors

[mnahkies]

What was changed

call the workflow interface dispose method, same as in github.com/temporalio/sdk-typescript/blob/main/packages/worker/src/workflow/vm.ts#L127-L130

see also: #1432

Why?

otherwise memory leaks occur when AsyncLocalStorage is used, as interceptors like

sdk-typescript/packages/interceptors-opentelemetry/src/workflow/index.ts

Lines 230 to 236 in 34ff54b

	export class OpenTelemetryInternalsInterceptor implements WorkflowInternalsInterceptor {
	async dispose(input: DisposeInput, next: Next<WorkflowInternalsInterceptor, 'dispose'>): Promise<void> {
	if (contextManager !== undefined) {
	contextManager.disable();
	}
	next(input);
	}

won't be run.

from my testing we re-evaluate modules on every workflow run, even when in the ReusableVMWorkflow is used. meaning eg: module level const AsyncLocalStorage instances are re-created for each workflow, and so it's important they are disposed of.

Checklist

1. How was this tested:
   Tested manually with a minimal example of our real workload. Pending testing with our production workflow code.

[CLAassistant]

All committers have signed the CLA.

[mjameswh]

Yeah, that definitely seems right. I'm totally baffled this bug didn’t get exposed earlier.

Thanks a lot @mnahkies for your work in investigating this.

[mnahkies]

Thanks for adding the test @THardy98 and getting this merged 🎉

Thought I'd just drop some runtime metrics showing the impact of this change for one of our workers that is under a roughly constant load of workflows, tldr; it is much more stable

• CPU now flat, rather than increasing with memory consumption
• Event loop delay now flat
• GC pause / count now flat

Previously it was OOMing frequently and restarting, with performance degrading up until the point it restarted.

This has also decreased the end-to-end duration of the workflows processed by this worker, roughly 2-3x improvement.