1) The noise and offset contributions of the rest of the opamp is attenuated by the gm(diff pair)/gm(M1) ratio. So, you wnat the diff pair gm large and M1 small.
2) You need good matching between M1 and its pair for good copy of the currents. So, no minimum L. If you look into the technology docs, sometimes matching is not even modeled for min L. They have too large mismatch. M4 is also copying the current from a reference (not shown), so you also need good matching there.
3) M2 & M3, you don't care. They are just cascodes, so, you can put them with min L.
Good luck