I am sure they did the math but at first I thought that landing legs would be unnecessary if you have a system that 'catches' the rocket on the ground. Since the rocket can hover, it would be possible to grab the rocket with a structure/claw.
The weight of the legs is probably quite low so this might have been the easiest solution.
A single Merlin engine, even at low throttle, has quite a bit more thrust than a (nearly) empty F9-1 stage has weight. Thus, it cannot hover; once it slows to 0, it quickly starts going up again.
Thus the landing profile is the "hover-slam": the engine is started at the exact point in time necessary to reach a "landable speed" (from 0-several m/s) at the ground, and is cut off when it gets to the target.
While modern avionics are quite good, it's still coming in pretty fast and from far away and will have some variance in location and speed, just because that's the real world. Probably they don't quite know how much yet, but it could easily miss the target by multiple meters and a few meters/s. And that's a heck of a condition to build some sort of giant grabbing apparatus for.
Perhaps once they're routinely landing on land, they may be able to determine they can hit the target well enough to build a grabbing contraption. Or maybe it's too hard. Or maybe they'd rather keep their infrastructure to a minimum: a concrete pad is a lot cheaper to build and easier to operate, and SpaceX wants to do a lot of flights.
Besides, the 4000lb of the legs is on a 40,000lb rocket.
I haven't checked the data, but being a pintle injector engine, there's potential for the Merlin to throttle quite deeply. Maybe they haven't done that yet.
It's published to be able to run at 70%, which is quite nice. But 70% is still way too much to match stage weight.
It's been said to be capable of 60% (or maybe 40%, depending on how you interpret a three-character tweet). But even that is still not enough: the stage is 40k lb empty and a Merlin D at sea level is 165k lb thrust. You'd need to get it to 25-33% to hover nearly-empty.
(Keep in mind most SpaceX numbers involve a fair amount of guesswork and Kremlinology, and they're always changing, but they should be ballpark accurate.)
This has been brought up multiple times in different places. The basic consensus is that a claw system would require more infrastructure on the ground as well as a more accurate guidance system.
The F9v1.1 cannot hover when landing as the TWR with an engine at 60% thrust (the minimum) is almost 2.
The weight of the legs is probably quite low so this might have been the easiest solution.