Improve the matching speed by limiting multiplicity

Now in matching we are doing all possible combinations of velo-segment and scifi-segment and then do a clone-removal, but it's really unnecessary, we can limit the max number of multiplicity, such as each scifi-segment can only match up to 4 best scifi-segments before clone-removal, this will speed up the algorithm without a significant drop in the efficiency.

FIY: @lohenry @jzhuo @dovombru @ascarabo @gligorov @cagapopo