Remove custom exit codes from Buildpack::on_error

[Malax]

Remove support for custom exit codes from Buildpack::on_error. Exit codes are part of the CNB spec and there are cases where some exit codes have special meaning to the CNB lifecycle. This put the burden on the buildpack author to not pick exit codes with special meanings, dependent on the currently executing phase. This makes Buildpack::on_error more consistent with the rest of the framework where we don't expose the interface between the buildpack and the CNB lifecycle directly but use abstractions for easier forward-compatibility and to prevent accidental misuse.

We actually made the mistake ourselves (#286) and had the default Buildpack::on_error implementation return 100 which means "failed detect" during detect instead of properly signalling an error.

Fixes #286

[edmorley]

Thank you!

I also wonder if at some point we should make libcnb_runtime_detect and libcnb_runtime_build stop returning exit codes of their own?

For the build case, it only ever returns 0, which seems better modelled by Result<(), ...>.

And for the detect case, it only ever returns 0 or 100 which seems better modelled by something like Result<DetectResult, ...> where DetectResult (or whatever it's called) is an enum with only two states?

Although that does mean that libcnb_runtime does end up having some CNB business logic in there (the mapping of those two detect enum variants), so maybe that's an argument against?

It just feels a bit bleugh to be returning i32s from libcnb_runtime_detect and libcnb_runtime_build.

[Malax]

I also wonder if at some point we should make libcnb_runtime_detect and libcnb_runtime_build stop returning exit codes of their own?

I have a new branch ready that at least removes all magic values for exit codes in the codebase. I'm not heavily opposed to the i32 values in those two functions since they're internal and just there to break up the code in libcnb_runtime. I think it becomes much better and (IMHO) good enough with constants.

EDIT: I just realised that both functions are no longer internal. They're pub, that might change things. I'll link the PR here as soon as it is ready. Ready now: #416