Improve scheduler coverage for AWS Batch and MPI edge/error paths

[Copilot]

This PR targets uncovered scheduler branches identified during coverage analysis, focusing on failure and boundary behavior that was not exercised by existing tests. It adds focused unit tests for AWS Batch array metadata/index handling and MPI scheduler initialization/work allocation edge cases.

• Coverage gaps addressed

  • Added explicit tests for AWSBatchArrayScheduler error paths:
    • missing arrayProperties.size in AWS Batch parent job metadata
    • non-integer AWS_BATCH_JOB_ARRAY_INDEX
  • Added explicit tests for MPIScheduler edge behavior:
    • mismatch between shared_vars and initializer return size
    • rank with no assigned items returns an empty index list

• Test isolation improvements

  • Added a resilient symbol loader in the new test module to import scheduler/constants directly from source when package-level imports are unavailable in lightweight environments.
  • Kept assertions specific to expected exception classes/messages to ensure the tests validate the intended branches.

• New test module

  • tests/test_scheduler_additional.py

Example of one of the added edge-case checks:

with pytest.raises(ValueError, match="invalid literal for int\\(\\)"):
    AWSBatchArrayScheduler().get_indices([1, 2, 3, 4])