This replaces most uses of LLDB's own String to int/float conversion implementation with
LLVM's to_integer. The only remaining code in LLDB that now still uses StringConvert is
debugserver which can't use LLVM code, so I'll leave that part for a future commit.
Unlike the other StringConvert patch this doesn't try to add any missing error handling
but just does a NFC refactoring to to_integer.
Consider inlining the call?