- Faster conversion to string for binary bases: linear instead of quadratic!

- Allocate smaller result for certain masking ops, e.g. (1L<<30000) & 1.