Issue #1385 demonstrates how a large 'precision' value can cause buffer
overflows. Originally, the buffer was designed to fit any scientific
notation. But the precision changes at which point large floating point
numbers are displayed in scientific notation or default notation.
In case of the default notation many extra zeros have to be printed,
this was not reflected in the output_buffer and an overflow could
occur.
This PR computes the number of zero that do not fit into the static
buffer and appends them at the end of the function triggering potential
a second dynamic allocation. (The first allocation is the std::string
allocation).
the convert_with_stringstream seems to be missing a
.imbue() call to set the locale independent of the global state.
Not setting this, can in some settings lead to errors, see Issue #1366.
