Avoid using floating point arithmetic and rely on safe_emalloc

Avoid using floating point arithmetic and rely on safe_emalloc
for the multiplication.

The actual size requirement is spelled out as:

** The result is written into a preallocated output buffer "out".
** "out" must be able to hold at least 2 +(257*n)/254 bytes.
** In other words, the output will be expanded by as much as 3
** bytes for every 254 bytes of input plus 2 bytes of fixed overhead.
** (This is approximately 2 + 1.0118*n or about a 1.2% size increase.)