This exposed a few underlying issues:

1. is_pod_struct was too strict to allow this. I've relaxed it to
require only trivially copyable and standard layout, rather than POD
(which additionally requires a trivial constructor, which std::complex
violates). Should the name be changed too? There might also need to be
some documentation to warn the user not to do something stupid with a
class that has invariants.

2. format_descriptor<std::complex<T>>::format() returned numpy format
strings instead of PEP3118 format strings, but register_dtype
feeds format codes of its fields to _dtype_from_pep3118. I've changed it
to return PEP3118 format codes. format_descriptor is a public type, so
this may be considered an incompatible change.

3. register_structured_dtype tried to be smart about whether to mark
fields as unaligned (with ^). However, it's examining the C++ alignment,
rather than what numpy (or possibly PEP3118) thinks the alignment should
be. For complex values those are different. I've made it mark all fields
as ^ unconditionally, which should always be safe even if they are
aligned, because we explicitly mark the padding.

GCC 4 doesn't provide std::is_trivially_copyable.

Require either a trivial copy constructor or a trivial copy assignment,
but not both. This is slightly more relaxed than being
TriviallyCopyable, which requires that both are either trivial or
deleted, and that at least one is not deleted. With this check, one of
them could be user-provided.

This moves the ^ in the format string (to specify unaligned) to
outside the `T{}` where it is sure to be parsed correctly. This is not
strictly necessary yet, but it paves the way for pybind#832.

For some reason, returning the 'value' constexpr char array as a string
would yield an empty string for format_descriptor<float>. I tried a few
variants of the constructor and this is the first variation I found that
worked.

At the request of @jagerman, a separate specialization of
format_descriptor is implemented for std::complex.

This may run into MSVC 2015 bugs again - will put back the workaround if
it fails.

This is to go with the weakened is_pod_struct static_assert.