compiler: Fix complex arguments and implement float16 lowering

[enwask]

Fixes some issues that prevented complex float arguments from being passed correctly. Also implements lowering for half-precision floats with the _Float16 type.

Some remaining tasks:

• MPI with complex floating types
• Look into MPI with float16, if possible
• Type conversion is done by modifying the global ctypes_vector_mappet; need some more tests for operator building/rebuilding/pickling with float16, or potentially a different (hopefully less brittle) solution
• More test coverage of existing behavior applied to float16 types

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[FabioLuporini]

but still needs tests

seconded, this is crucial

[enwask]

Yeah my plan was to write some more extensive tests for complex (since the existing ones didn't catch all of the break cases I've found) and add in a few for half. I want to get all of these changes out of the way first, and then there's a few more things to work out such as math functions being assigned to float16 arrays. Maybe this should be a draft actually?

[mloubout]

Merged into #2375