Increase file descriptor limits on ECS/Fargate workers (#174)

* This change resolves the issue of the maximum number of open
  file descriptors being exceeded at runtime. This is documented
  in the Dask Distributed FAQs
  (https://distributed.dask.org/en/latest/faq.html). The issue
  occurs on FargateCluster nodes under some types of heavy
  workloads.
* The `nofile` hard and soft ulimits are set to 65535 (from default
  1024) in both the scheduler and worker nodes via the ECS task
  definitions for each.
* Also see dask/distributed#1858

dask_cloudprovider/aws/ecs.py

@@ -1105,6 +1105,13 @@ async def _create_scheduler_task_definition_arn(self):
                             if not self._scheduler_extra_args
                             else self._scheduler_extra_args
                         ),
+                        "ulimits": [
+                            {
+                                "name": "nofile",
+                                "softLimit": 65535,
+                                "hardLimit": 65535,
+                            },
+                        ],
                         "logConfiguration": {
                             "logDriver": "awslogs",
                             "options": {
@@ -1171,6 +1178,13 @@ async def _create_worker_task_definition_arn(self):
                             if not self._worker_extra_args
                             else self._worker_extra_args
                         ),
+                        "ulimits": [
+                            {
+                                "name": "nofile",
+                                "softLimit": 65535,
+                                "hardLimit": 65535,
+                            },
+                        ],
                         "logConfiguration": {
                             "logDriver": "awslogs",
                             "options": {