XML ubsan fixes

[cmazakas]

The XML oarchive class generates a few ubsan failures, this PR aims to fix them.

To replicate the failures on develop:

b2 -q test//test_mult_archive_types cxxstd=20 undefined-sanitizer=norecover address-sanitizer=norecover  link=static toolset=clang-16

The first failure is that we're incrementing a nullptr in the increment() of the xml_escape iterator because we never assigned m_bnext to non-null.

The second is a static_cast to a Base that we just replace with a reinterpret_cast, trying to utilize the common initial sequence to silence the sanitizer.

We found this in Unordered when adding Serialization support to some of our containers.

[pdimov]

This change is broken and causes crashes.

[pdimov]

.\boost/archive/detail/interface_oarchive.hpp(55,16): warning: 'reinterpret_cast' to class 'boost::archive::text_oarchive *' from its base at non-zero offset 'boost::archive::detail::interface_oarchive<boost::archive::text_oarchive> *' behaves differently from 'static_cast' [-Wreinterpret-base-class]
        return reinterpret_cast<Archive*>(this);
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.\boost/archive/detail/interface_oarchive.hpp(78,15): note: in instantiation of member function 'boost::archive::detail::interface_oarchive<boost::archive::text_oarchive>::This' requested here
        this->This()->save_override(t);
              ^

[pdimov]

testing.capture-output bin.v2\libs\msm\test\SerializeSimpleEuml.test\msvc-14.2\release\cxxstd-latest-iso\threading-multi\SerializeSimpleEuml.run
====== BEGIN OUTPUT ======
Running 1 test case...
unknown location(0): fatal error: in "my_test": memory access violation occurred at address 0x4, while attempting to  read inaccessible data
libs\msm\test\SerializeSimpleEuml.cpp(134): last checkpoint

*** 1 failure is detected in the test module "MyTest"
 
EXIT STATUS: 201 
====== END OUTPUT ======

[robertramey]

I was sort of skeptical of this change. But I didn't have the time/inclination to really investigate it so I merged it. On my local machine All tests still passed. I'll change this back to the way it was.

[robertramey]

Done!

[cmazakas]

Yeah, I'd like to apologize for not vetting my changes more carefully.

Initially, I was overwhelmed by the size of Serialization and how it was laid out.

Going forward in the future, I'll test changes locally a bit more carefully before opening another PR.