Drain any remaining activations from `poll_task` until it returns `CoreError::ShuttingDown`

[bergundy]

Doesn't need to be this PR but you'll probably want to have some mechanism to drain any remaining activations from poll_task until it returns CoreError::ShuttingDown. Remaining activations may exist due to replaying a workflow to catch up to the end of history which was already fetched from the server

Originally posted by @Sushisource in #18 (comment)

[bergundy]

@Sushisource I tried to implement draining but never got CoreError:ShuttingDown, in fact the promise returned from run was never resolved or rejected.
See the worker code here: https://github.com/temporalio/sdk-node/blob/aff47fddc0f2de53f52677fc97084139ea268e06/packages/worker/src/worker.ts#L142

Example worker to reproduce:

import { Worker } from '@temporalio/worker';
import { sleep } from '@temporalio/worker/lib/utils';

const shutdownGraceTimeSecs = 5;

async function run() {
  const worker = new Worker(__dirname);
  process.on('SIGINT', async () => {
    console.log('Shutting down worker');
    worker.shutdown();
    await sleep(shutdownGraceTimeSecs * 1000);
    console.error('Timed out while waiting for worker to shutdown gracefully');
    process.exit(0);
  });
  await worker.run('test');
  console.log('Worker gracefully shutdown');
}

run().then(
  () => void process.exit(0),
  (err) => {
    console.error(err);
    process.exit(1);
  }
);

[Sushisource]

@bergundy I'll fix this - it's immediately clear to me what the problem is... poll_task is stuck blocked on the long poll and never gets a chance to see that the shutdown flag is set. I need some way to interrupt the long poll when that happens. Should be easy.

[Sushisource]

Should be fixed by temporalio/sdk-core#59

[bergundy]

After the above is merged, we should update the submodule and create a PR for https://github.com/temporalio/sdk-node/tree/graceful-shutdown