Fix potential "infinite recursion" panics in runners

[ag-eitilt]

We've seen a couple problems where the proper reporting of an issue got masked by an "Infinite recursion detected" panic, and we verified that it was due to primJobFinish/primJobFailFinish being called before the respective primJobLaunch/primJobFailLaunch -- which for all their other benefits is a new risk with v44's flattened runners. Ultimately it would be best to implement an architecture solution for solving this, but in the mean time it's sufficient to patch the existing runners and to keep a careful eye on future code which touches them.

[colinschmidt]

One question about the mkdir removal, otherwise LGTM

[colinschmidt]

Where does this mkdir live now?

[ag-eitilt]

Inside the write:

export def write (path: String) (content: String): Result Path Error =
    def spath = simplify path

    require Pass parent =
        simplify "{spath}/.."
        | mkdir # Note that this is recursive, as well.

    writeImp (parent, Nil) 0644 spath content

My guess is that the mkdir was originally split out purely for the error message, but the shell's mkdir already advertises the path it tried unsuccessfully to create, so keeping it separate didn't seem to have any benefit.

Also, CI apparently succeeded? I'll look into the supposedly end-of-lifed image in a bit.

[ag-eitilt]

Ah, apparently the deprecation is happening on jan 30, so we've got another month and a half. Merging!