Clean up jobs periodically or on some trigger

[zjrgov]

There are a few issues that this would solve:

• Cancelling a job in the GitLab UI sometimes creates zombie workers & services
• Pipelines set to disable cleanup are run for debugging purposes and require manual cleanup. (see Flag to Preserve Workers/Services for Debug #39, and Add flags to keep GLR-W/S from being cleaned #40)

We'll need some mechanism to run cleanup when the runner gets a cancellation signal and to run clean up periodically.

[zjrgov]

We should make sure to listen for SIGTERM (and maybe SIGKILL): https://docs.gitlab.com/runner/executors/custom.html#terminating-and-killing-executables.