[libc][NFC] Fix build when using UBSan

[abrachet]

Summary

There is a problem in memory_utils where adding an offset to a nullptr triggers ubsan (nullptr was used in tests for simplicity). As well as a problem in MPFRNumber when width is specified as 0 causing bit shift that was out of bounds.

Hopefully these changes are obvious but wanted to make sure.

[sivachandra]

I am OK with this but added @gchatelet so that he can be aware of this.

[gchatelet]

Is this still to be submitted?

[abrachet]

It looks like I never committed this, no. Just a quick look in MPFRUtils.cpp now, it has changed substantially that I couldn't find where this shift is happening anymore. I could look more into this over the weekend

libc/test/src/string/CMakeLists.txt

 add_libc_testsuite(libc_string_unittests)
 
+add_compile_options("-fsanitize-recover=pointer-overflow")
+
 add_subdirectory(memory_utils)
 
 add_libc_unittest(
   strcat_test
   SUITE
     libc_string_unittests
   SRCS
     strcat_test.cpp

libc/utils/MPFRWrapper/MPFRUtils.cpp

public:
   }
 
   template <typename XType> MPFRNumber(XType x, const Tolerance &t) {
     mpfr_init2(value, mpfrPrecision);
     mpfr_set_zero(value, 1); // Set to positive zero.
     MPFRNumber xExponent(fputil::FPBits<XType>(x).getExponent());
     // E = 2^E
     mpfr_exp2(xExponent.value, xExponent.value, MPFR_RNDN);
-    uint32_t bitMask = 1 << (t.width - 1);
+    uint32_t bitMask = t.width ? 1 << (t.width - 1) : 0;
     for (int n = -t.basePrecision; bitMask > 0; bitMask >>= 1) {
       --n;
       if (t.bits & bitMask) {
         // delta = -n
         MPFRNumber delta(n);
 
         // delta = 2^(-n)
         mpfr_exp2(delta.value, delta.value, MPFR_RNDN);