[WIP] MemorySanitizer and UndefinedSanitizer builds

[mosra]

Things to do:

• Fix all "null src" errors in std::memcpy() ((UndefinedBehaviorSanitizer) GrowableArray.h: null pointer passed as argument 2, which is declared to never be null  #102)
• Fix all null pointers errors in std::memcmp() -- String, StringView (ffs? really? come on, C)
• Apparently std::memchr() also can't accept null input, WTH

Original (most probably obsolete) TODO below:

• The MemorySanitizer needs libc++, which is impossible to get on Travis:
  • It is in the whitelist for Precise, but just adding libc++-dev to package list on Precise gets just ignored, because there's no such package the default repos and apt-get treats not found packages as regular expressions, which then matches libc6-dev or about anything else. So the installation succeeds, but silently doesn't install the package at all, failing during compilation with can't find -lc++. Whoever thought this is a good idea for a package manager?!!
  • It is not in the whitelist for Trusty.
  • The only PPA that has libc++-dev for Precise is debian-sid. Just enabling that PPA makes all hell break loose and apt-get install fails on some file conflict much later.
• The UndefinedSanitizer reports weird errors about mismatched class inheritance (false positive?) and apparently UBSan itself leaks, as it triggers a error from ASan on exit. There are also a bunch of other reports that need an investigation. See the build output for details.

[mosra]

Totally by accident I found some info on the UBSan false positives related to vptr with dlopen(): https://whatofhow.wordpress.com/2015/03/17/odr-rtti-dso/

It references these commits, which look like they export polymorphic classes and load the dynamic objects using RTLD_GLOBAL:

• https://cgit.freedesktop.org/libreoffice/core/commit/?id=f0aa1a78fb209310e8baef53c02f365fca518d11
• https://cgit.freedesktop.org/libreoffice/core/commit/?id=b4f6b26b5a1a78fecfa95ec2eb7ac8b80495d8aa

But I am already doing all of that? Another thing is using -fvisibility=ms-compat instead of -fvisibility=hidden, so maybe that's the key.