SDK harnesses should be able to indicate panic in FnAPI

If the SDK harness encounters a permanent problem (local file corruption or bad staged files, say) it should be able to rely that information to the runner as a "panic" message, say. Such conditions may also happen during the boot stage.

Using process/container exit codes would likely not work well, because that information may not flow easily to the runner logic depending on how containers are managed.

Imported from Jira [BEAM-2947](https://issues.apache.org/jira/browse/BEAM-2947). Original Jira may contain additional context.
Reported by: herohde.

Start with the FnAPI SDK harness and runner handling described in the issue, including boot-stage behavior, and review the imported Jira BEAM-2947 for the missing context. Define how a permanent harness failure is represented as a panic message and how the runner receives and handles it.

Written by the indexing model from the issue text.