…rrat-pvf-fork-instead-threads

…rrat-pvf-fork-instead-threads

…rrat-pvf-fork-instead-threads

…rrat-pvf-fork-instead-threads

…rent

- Changed error for pipe read failure (job died)
- Updated seccomp detection (IoErr no longer means job died)
- Updated test
- Confirmed manually that a seccomp violation (with process killing) result in a
  JobDied error.
- Updated some other errors (still more to do)

…at/polkadot-sdk into jpserrat-pvf-fork-instead-threads

…ads' into jpserrat-pvf-fork-instead-threads

- Created a `JobResponse` and `JobError` types for the job to self-report its
status. Note that these reports must be treated as untrusted due to the
job running untrusted code. Previously some errors from the job were being
treated as internal! That's why it's important that the error handling here is
robust and accurate.

- Made `WorkerResponse::JobError(String)` hold any unexpected error from the job
process.

- Renamed some of the response types and variants to be more accurate.

There were several simultaneous issues:

- We were using the wrong exit calls in the allocation failure handler. This is
some evil stuff that I don't understand.

- We were ignoring responses from the child when the exit status was 1

- We were sending a response encoded with a length prefix, but not decoding with
`framed_recv`