Remove the Windows ROS3 shutdown-deadlock CI workaround once HDF5 fixes it upstream

[rly]

The Windows ROS3 CI jobs currently run through a workaround for an upstream HDF5 defect. This issue tracks removing that workaround once the defect is fixed.

Background

• Symptom: Windows ros3 test jobs hang on process exit (since 4.0.0) #2228 (Windows ROS3 test jobs hang on process exit)
• Workaround: Work around HDF5 ros3 shutdown deadlock on Windows CI #2229
• Upstream defect: ROS3/S3 VFD deadlocks at process exit on Windows HDFGroup/hdf5#6560 (ROS3/S3 VFD deadlocks at process exit on Windows)

When to act

When an HDF5 release containing the fix is available in the conda-forge builds that the ROS3 CI jobs install.

What to back out

• Delete scripts/test_ros3.py.
• Restore the bottom of test.py to sys.exit(main()), dropping the PYNWB_EXITCODE_FILE handoff.
• Point the four ROS3 job steps in .github/workflows/run_tests.yml and .github/workflows/run_all_tests.yml back at python test.py --ros3 and python test.py --example-ros3.
• Decide whether to keep timeout-minutes: 20 on the two run_all_tests.yml ROS3 jobs. It is a useful backstop independent of this bug.
• Confirm the Windows ROS3 jobs exit cleanly on the scheduled Run all tests workflow.