I've been following this topic as a curiosity. I'm not totally certain but I think the LM systems don't have adaptive optics. But DARPA had a program where they were making something analogous to a phased array out of fiber lasers, where each individual laser had its own independent lens, and a phase shifter is used on each to align all the beams in real time so they're all mutually coherent. The phase shifter is tuned in real time by reflection back along each beam line from the target, split off with a beam splitter to some sort of sensor.

They stopped talking about it after like 2 years, which maybe means it was a flop or maybe they took it secret. Who knows. If you're curious this is the name to google: https://www.darpa.mil/program/excalibur

Can lasers be steered with phase shifting as radar can?

Not to the same extent. Each element in a radar phased array has a wide angle of emission, and the overlap is necessary to produce the final aggregate steered beam. Lasers need to have as low dispersion as possible to be useful at range, so they can't do that. The project above used some other means of pointing for broad beam steering. The phase shifters are just such that each individual laser is closer to constructive vs destructive interference at the target surface.

If your beam is perfectly coherent, then its cross section has the same area regardless of the distance, so you should be getting full power. However, no beam is perfectly coherent, so it’s still O(1/r^2), and the better coherence, the smaller constant you have hidden in the big-O notation.

1/r^2 comes from the fact that space is three dimensional. No amount of beam collimating can change that (unless we have some fundamental breakthrough/change in our understanding of physics). Better beam coherence could keep improving on the solid angle, which is the multiplicative coefficient to the (asymptotic) scaling behavior.