Triage integration tests that fail on windows

[philipcraig]

#817 shows that several integration tests fail on the nightly Windows msvc or gnu rustup channels

Some of the causes of the failures are:
test_stringview uses string_view. This requires C++17, and the flag for this is not supplied to the Windows compilers
test_forward_declaration destructs a unique_ptr of an incomplete type, which is non-compliant C++ #840
test_bitset contains Clang-specific __type_visibility__ attributes
test_protected_inner_class uses designated initialisers which is only compliant for C++20
test_private_inner_class uses designated initialisers which is only compliant for C++20

Other test failures are reproducible, but so far untriaged

[adetaylor]

A (presumably) flakey test: https://github.com/google/autocxx/runs/5337017794?check_suite_focus=true - test_uniqueptr_moveit. This might be a real bug which just happens to manifest on Windows only due to ABI issues. I'd like to figure out how to run the test suite under ASAN on Linux to see if any problems show up.

[philipcraig]

A (presumably) flakey test: https://github.com/google/autocxx/runs/5337017794?check_suite_focus=true - test_uniqueptr_moveit. This might be a real bug which just happens to manifest on Windows only due to ABI issues. I'd like to figure out how to run the test suite under ASAN on Linux to see if any problems show up.

It's unusual that it's flaky. None of the already ignored Windows tests were flaky. The flakiness in tests that we saw while I was adding channels was in the beta and nightly channels on all of the architectures, due to incremental compilation bugs. There of course could be more of those.