Linear issue: ENG-5177

This error is not really causing any problems. That said, its appearance can be considered a regression caused by D9298.

Before that diff, the check on this.ws.readyState at the start of sendMessage prevented the error. After that diff, there was a possibility of this.ws.readyState changing after the compression step.

This diff addresses the regression by introducing a second check after the compression step.

To be honest, this would be fairly expensive to test compared to the cost of making the diff, and the priority of the issue it's addressing. I think the diff is fairly safe... relying on Flow and CI to catch any issues.