Remove cpp11_should_unwind_protect altogether

[DavisVaughan]

Closes #325
Closes #326

• Update documentation to talk about dangers of nested unwind protection
• Update documentation to talk about dangers of using C++ code that calls destructors from within a manual unwind_protect() call (since the destructors won't run if a longjmp occurs)
• Update documentation to talk about performance of cpp11 character vectors in a loop, and how to use unwind_protect() manually to avoid the performance hit
• Update NEWS bullet
• 2nd order revdep checking (these came back clean)

Note that this makes the example from #298 very slow again, because nested unwind_protect() is no longer "optimized". This is an example where character vector manipulation may be better with a manual unwind_protect() + R C API usage

[kevinushey]

LGTM!

[lionel-]

Nice writeup!