[WIP] Split `arb_pow_fmpz_binexp` into a `ui` and `mpz` version and implement proper `arb_sqr`

[albinahlback]

I am done with the former part, with addition of adding nn_sqr_2 which simplifies nn_mul_2x2 by skipping one umul_ppmm.

I am going to add arf_sqr_rnd_down and then implement arb_sqr from there.

[fredrik-johansson]

Did you do some profiling?