The number of windows is determined by the required force constant, which in turn depends on the shape of the PMF. You may have to do a trial-run to see that your histograms are overlapping well enough.
It is difficult to give any specific recommendations on the sampling time for each window. It depends on the equilibration time (in each umbrella window) and the pull force autocorrelation times in each window. Different windows might have different “optimal” sampling time. Regarding the equilibration time (the time to discard in the beginning of each umbrella window using gmx wham -b) you might find more information in Bumpy energy profile from PMF - #4 by MagnusL and the following posts. Apart from that, you can often see if there is a large bootstrap error in specific regions in your PMF. Those regions might require more sampling time.