1) Is the baseband approach of frequency synchronization only of theoretical interest or are they actually used in products
With the development of full-digital receiver, now the baseband frequency synchronization is already used in products.
2) Is the synchronization task performed in two stages? eg- first at RF/IF level and then remaining offset in baseband software
Two stages are better. But only one stage is also acceptable. It depends on the signal characteristics.
3) How is synchronization achieved in USRP- at daughterboard hardware or gnuradio software
On the daughterboards, there is no synchonization processing. Everything is made in gnuradio software. You may use 'soft PLL' or 'soft baseband SYNC'.