Use prefix increment in make_iterator

[aldanor]

This patch switches to using prefix increment for iterators returned by make_iterator. This is generally a good idea since postfix increment involves creating a temporary object which may sometimes be expensive.

Btw I've just noticed there's been another related PR (#183) which seems to have been abandoned.

[wjakob]

Both implementations (pre and postfix) will make a copy of the object (rather than returning a temporary), since the return type of the lambda function is ValueType. When the iterable stores pointers, this means a copy of the pointer (i.e. no costly copy of the pointee).

(So it's not clear to me how this is different/better than the prior approach -- can you clarify?)

[aldanor]

@wjakob

1. The assumption that copying pointee is always cheaper than pointing the iterator is not always true. I'm currently working on a codebase where the yielded value type is simple (like an int), but the iterator itself contains a shared_ptr around a handle that wraps a C-level FFI pointer. This way, copying the iterator is more costly and is completely uncalled for.
2. pybind11 is the first library I've encountered (STL, Streams, cppitertools, a bunch of others) that requires implementing postfix++, anywhere else (including language constructs like range for, etc) just implementing a prefix increment is entirely sufficient.

[wjakob]

ok -- I think those are good reasons. Thanks!