@batch step fails locally with 'No IAM role specified' when no per-job IAM role is configured

[aspratyush]

On EC2 compute environments that use an instance profile for container credentials, running any @batch step fails immediately — before the job reaches AWS — with:

BatchJobException: Unable to launch AWS Batch job. No IAM role specified.

METAFLOW_ECS_S3_ACCESS_IAM_ROLE is not set; the container is expected to acquire credentials from the EC2 instance profile.

Metaflow version: master, 4871fb1
Deployment: AWS Batch on EC2 compute environment, no per-job IAM role

Steps to reproduce

1. EC2 compute environment using instance-profile credentials
2. METAFLOW_ECS_S3_ACCESS_IAM_ROLE not set
3. Run any @batch-decorated flow

Expected
Job submits successfully; the AWS Batch ContainerProperties API documents jobRoleArn as optional.

Actual
The guard in batch_client.py (L82–85) raises unconditionally when iam_role is None, and _register_job_definition always passes "jobRoleArn": job_role to boto3 even when job_role is None, which boto3 rejects for string fields.

[odncode]

I'd like to work on this. The fix looks straightforward, make jobRoleArn conditional in _register_job_definition and remove the unconditional raise when iam_role is None, since the AWS Batch API documents jobRoleArn as optional.

[odncode]

The fix for this is implemented in #3211 and CI is now green.

Happy to discuss the approach whenever a maintainer has the bandwidth.

Short version: jobRoleArn is only included in the job definition when a role is actually configured, since boto3 rejects None for that field.