[libcxx] Explicitly cast in generate_canonical

[bcain]

Summary

A new clang warning introduced in r367497 was complaining about
the change in value.

[bcain]

This patch is intended to address new warnings like the one below that shows up with clang on a hexagon target:

libcxx/include/random:3650:40: error: implicit conversion from 'unsigned long' to 'float' changes value from 2147483645 to 2147483648 [-Werror,-Wimplicit-int-float-conversion]
    const _RealType _Rp = _URNG::max() - _URNG::min() + _RealType(1);
                          ~~~~~~~~~~~~~^~~~~~~~~~~~~~ ~

I *think* the behavior is intended/expected, so that's why I include explicit casts here to avoid the implicit conversion warning.

The test case has corresponding warnings and I think I'll just include those in this commit here.

[ldionne]

This looks right to me (and after a short discussion with Marshall it seemed alright to him too). I'm going to check it in to fix the several builders that are broken right now.

libcxx/include/random

generate_canonical(_URNG& __g)
     const size_t _Dt = numeric_limits<_RealType>::digits;
     const size_t __b = _Dt < __bits ? _Dt : __bits;
 #ifdef _LIBCPP_CXX03_LANG
     const size_t __logR = __log2<uint64_t, _URNG::_Max - _URNG::_Min + uint64_t(1)>::value;
 #else
     const size_t __logR = __log2<uint64_t, _URNG::max() - _URNG::min() + uint64_t(1)>::value;
 #endif
     const size_t __k = __b / __logR + (__b % __logR != 0) + (__b == 0);
-    const _RealType _Rp = _URNG::max() - _URNG::min() + _RealType(1);
+    const _RealType _Rp = static_cast<_RealType>(_URNG::max() - _URNG::min())
+      + _RealType(1);
     _RealType __base = _Rp;
     _RealType _Sp = __g() - _URNG::min();
     for (size_t __i = 1; __i < __k; ++__i, __base *= _Rp)
         _Sp += (__g() - _URNG::min()) * __base;
     return _Sp / __base;
 }
 
 // uniform_int_distribution