Fix longjmp marshaling tests, clean up relifting

[sim642]

Besides just fixing the necessary, this does broader cleanup with the hope to avoid similar issues again.

Changes

1. Remove identity Printable.Std.relift which hides missing definitions.
2. Add all missing relift definitions. Some rather core types like MapDomain were not relifting their inner components.
3. Introduce Printable.StdLeaf as Printable.Std alternative with identity relift. This is only to be used with primitive and CIL types which don't contain inner types that could require relifting.
4. Remove default Printable.Std.name, which names everything "std". This makes debugging such issues a total pain, e.g. "lifted std and either std or std option" is completely ineligible. All printables must specify a name explicitly.
5. Add all missing name definitions.

TODO

• Readd relifts for all other marshaling tests.

[michael-schwarz]

Might be worth doing some of the incremental benchmarks here, if this is very expensive, it might be worth taking steps to avoid relifting where it is not needed due to the current configuration.

[michael-schwarz]

Also we might think about writing a deriver for relift? For most cases I don't see a reason we cannot do it.

[sim642]

Might be worth doing some of the incremental benchmarks here, if this is very expensive, it might be worth taking steps to avoid relifting where it is not needed due to the current configuration.

That could be rather error-prone again though. If all possibilities for sub-printables are not accounted for in a particular configuration, it leads to similar hard-to-debug issues.
Recursing more might cause us to do extra work, but overall this would be linear in the size of the data. Traversing that in memory shouldn't be slower than reading it from a file.

Also we might think about writing a deriver for relift? For most cases I don't see a reason we cannot do it.

I'll add it to #31. I started working on it at some point, but it's been in hibernation for a long time. I wanted a more general and maintainable approach than writing a dozen ppx derivers which duplicate large parts of the code just for a different function name/return type/number of arguments/etc.