Added support for sending blocks to objc

[ojii]

This adds support for sending blocks to objc. It supports both "automagical python callable to objc block style" (which in non-trivial cases can segfault due to python GC) as well as "explicit create a block object style" through the rubicon.objc.Block.

Magical style (functions need to be fully annotated):

def myblock(arg1: arg1type, arg2: arg2type) -> return_type:
    ...

objc_instance.ObjcMethod_(myblock)

Explicit style:

from rubicon.objc import Block

myblock = Block(lambda a, b: ..., return_type, argtype1, argtype2)
objc_instance.ObjcMethod_(myblock)

[dgelessus]

Thanks for the PR! Overall everything looks good to me, but I have a couple of small questions/comments, see below.

Sorry that I haven't been able to review this sooner, I wasn't at my computer for the last few days.

[dgelessus]

The __NSGlobalBlock__ class isn't actually documented anywhere. _NSConcreteGlobalBlock is documented, but not required to be an Objective-C class (you're only supposed to use its address).

So the officially supported way to use this would be:

_NSConcreteGlobalBlock = (c_void_p * 32).in_dll(c, "_NSConcreteGlobalBlock")
# ... when setting the isa ...
self.literal.isa = addressof(_NSConcreteGlobalBlock)

(_NSConcreteGlobalBlock is declared as a void *[32] in Block.h.)

[dgelessus]

This discards the return value from the encoding_for_ctype call - is this intentional? If so, it would be good to add a comment.

[ojii]

looks like leftover debugging code from me.

[dgelessus]

This should probably be in the except KeyError block below, so that pointer types that are in _encoding_for_ctype_map are looked up there. For example POINTER(c_char) should encode as *, not as ^c.