Fixed typo in _castf primitives on Z80 that lead to linker errors

[danielgaskell]

The software floating-point library used on Z80 contains multiple primitives with double underscore prefixes, like

long __castf_l(uint32_t a1);

This compiles to three underscores in the object file, when the compiler only expects two, leading to "unknown symbol" errors when casting floats. Easy fix.

[danielgaskell]

(Actually, this looks like a problem with several other targets as well.)

__pluseqf etc. primitives are also missing.