Fix compilation with gcc < 5

[mvoelkle-cern]

Description

When the user defines _GLIBCXX_USE_CXX11_ABI=0 to force the pre-c++11 ABI, numpy.h assumes that is_trivially_copyable is available.
It is not necessarily the case. This patch uses clang's feature detection instead.

Suggested changelog entry:

Fixed compilation with GCC < 5 when the user defines ``_GLIBCXX_USE_CXX11_ABI``.

[EricCousineau-TRI]

Howdy @mvoelkle-cern! Do you have a theory as to why our present CI coverage doesn't already check this case?
From a naive read at present master, it seems like we should have?

pybind11/.github/workflows/ci.yml

Lines 565 to 574 in 114be7f

	# Testing on CentOS (manylinux uses a centos base, and this is an easy way
	# to get GCC 4.8, which is the manylinux1 compiler).
	centos:
	runs-on: ubuntu-latest
	strategy:
	fail-fast: false
	matrix:
	centos:
	- 7 # GCC 4.8
	- 8

Also, I believe these CI errors are caused by your PR - can you fix those?

[mvoelkle-cern]

Hi, sorry my fix didn't compile outside of clang, I had wrongly assumed that preprocessor boolean expressions would short-circuit. Here is an other approach that targets libstdc++ specifically.

[mvoelkle-cern]

Howdy @mvoelkle-cern! Do you have a theory as to why our present CI coverage doesn't already check this case?
From a naive read at present master, it seems like we should have?

pybind11/.github/workflows/ci.yml

Lines 565 to 574 in 114be7f

	# Testing on CentOS (manylinux uses a centos base, and this is an easy way
	# to get GCC 4.8, which is the manylinux1 compiler).
	centos:
	runs-on: ubuntu-latest
	strategy:
	fail-fast: false
	matrix:
	centos:
	- 7 # GCC 4.8
	- 8

The issue is when using 4.8 AND the user defines _GLIBCXX_USE_CXX11_ABI=0. The CI coverage doesn't define it.

[EricCousineau-TRI]

I can't state w/ 100% confidence that I can verify, but will trust testing on your side as well as our CI (and of course, Henry's review as well). Thanks!