To the original question: Running side fills is a little less intuitive than it seems. It is often referred these days as a cross fill. This is because your not sending a stereo copy of your FOH mix to those on stage. You sending the musicians a mix that would be more like what exists on the other side of the stage. I.E. You would send stage left instruments into the stage right side fill and visa versa. The idea behind the side fill is to bring things back into the mix that are missing on that side of the stage. A guitar rig on SL will not be as apparent on SR. So you add that guitar into the SR side fill to bring him into the mix more. The idea is to balance the sound so the whole band feels like they are in front of the PA.
As for polarity of the monitors and cross fills. That is dependent upon which will yield the highest gain before feedback? Usually normal polarity will work just fine. If there seems to be weird stuff going on, you can try flipping the polarity for improvements. The issue is usually not the speakers as much as it's the 16+ mics that are opened up across the stage. You can get almost as much comb filtering from the speakers as you can from the mics. This is especially true if the band has a particularly loud stage level. Loud stage levels usually mean loud monitors and all bets are off when you have more volume going backwards than you have going forwards.
As for subs to tops polarity relations, they should be lined up. In many cases doing a null test will help determine the best setting. Turn off one side of the PA and test with only one side at a time. Send an 80hz-100hz ( your choice depending on your crossover location ) signal into the tops and subs. Change the polarity on the tops to get the highest drop in level of the tone. This lets you know that the two elements are out of phase. Now you want to align the drivers. You can do this by using electronic delay, or physically moving the speakers. You want to null out as much of the 80hz signal as possible. So either add delay to the tops, or subs for the largest level drop, or physically move the speakers back and fourth for the largest level drop. Once you have nulled out as much of the 80hz signal as you can, flip the polarity back on the tops and you should now have maximum summation between the tops and the subs. Repeat this for the other side and then turn everything on together and listen to see that both sides are in polarity. If they are out it will be apparent right away.
More on speaker cancellations and all that crap. Any time you have the same sound coming out of two sources at the same time you will have cancellations, or comb filtering as it called. However if the level and the physical time is different there will be less cancellations. In the case of monitors, which should be lower in level, has a different arrival time and are technically out of phase anyway, there will be little to no major impact on sound other than in the lower frequencies. Being that most monitors are rolled of below 80hz there is no real impact except in the center octaves. This can be large or small dependent upon the distance between the FOH and monitors. The further upstage from FOH the monitors are, the lower their impact. As they become closer in line, such as the case on smaller, tighter stages, cancellations can become more noticeable.
If you take two speakers and place them next to each other pointing in different directions, you only have cancellation, or addition in the areas in which the coverage overlaps. Otherwise they will project as normal. If you take two speakers and place them at 90 deg. in relation to each other facing in the same general direction. You will have addition where they cross and very little will be affected beyond that. Now if they were pointing in opposite directions you will only have cancellations when you are between the two speakers, which if properly deployed for the circumstance ( like on either side of a fence or a tent wall ) should yield no ill effect for the intended listening area.