2) Sure, it's ironic to harness solar energy to give light to plants, but it might well be more effective. Plants can in a theoretic perfect world collect 11% of the sun's energy for use in photosynthesis according to Wikipedia. In effect the figure is more like 3% to 6%. A majority of solar panels have an efficiency of between 15-17%. The building's owner might actually be able to sell surplus energy to the market after lighting and warming/cooling their plants. That's not mentioning wind power which is readily available at the top of tall structures. As for the lighting, I'm not sure where your claim that you're losing 90% of energy comes from... LEDs are quite effective.
3) As I've pointed out above, your dismissal of this idea rests on quite a few incorrect assumptions. Shipping is cheap..ish.. but there's an opportunity cost associated with it. Hong Kong, for example, is extremely reliant on imports of food. 47% of the city's fruit go through one market, 317,000 tonnes of it in 2014. If just a portion of that, by weight, was taken off the streets by urban farming on brownfield plots, rooftops, or property that for regulatory reasons can be neither residential or industrial (noise/safety) the cost of shipping other goods would drop.
I'm glad you mention the water. By 2050 the World Bank estimates the number of urban dwellers with seasonal water shortages will reach 1.9bn people. If urban farming can help this issue, even if all the other benefits weren't present, it would be worth it.
So the sunlight->panel->led->light conversion does waste ~90% of the sunlight.
This is made less bad by not lighting plants with white light but purple light, but clearly solar panel on the roof, plants indoor don't make much sense.
Now, the reasoning is completely different if the electricity is produced elsewhere, transporting produce may not be that expensive, but transporting electricity is a lot cheaper.
The main benefit from urban farming is definitely reduced time from harvesting to the plate, and the ability to consume varieties that don't transport well...
Why not? Wasting 90% of the sunlight is still better than using sunlight directly, where the plants themselves only use 4-7% of the sunlight. The plants are wasting all the green-spectrum light, which is where most of the sun's energy is.
Besides, you wouldn't use white LEDs, there's no sense in that. You'd use LEDs tuned for the wavelengths that plants actually use.
At a cost of atleast $2 a per watt for lights (possibly up to $4 a square foot for industrial/commercial electrical rating) and assuming 35 watts per square foot, a 2000 square foot grow space, you are looking at $140,000 in LEDs alone. plus 70KW for 16 hours a day(+10% more for supply inefficiencies), that's about 1200 kilowatt hours of power needed to supply daily. Not sure how many panels that is but it has to be a lot.