First Cap footprints: Main cap, 35 to 50V rating, requires 1210 (or 1206) specifically this:
http://www.digikey.com/product-detail/en/samsung-electro-mechanics-america-inc/CL31B106KBHNNNE/1276-6767-1-ND
Maybe use the same cap before the FET.
If you can fit 22mm on the Q8 I definitely would. You can bring 26mOhm down to 7 (BIG efficiency savings) and get a full 15A rating!
http://www.digikey.com/product-detail/en/wurth-electronics-inc/74435582200/732-4237-6-ND
Worry about smaller ones for smaller drivers. We’ve found a few, easy to come back to them. Further I would add through holes to allow optionally mounting unshielded toroid inductors unless there’s really no place for them to exist. It will give options for testing/learning/improving and that might help down the road.
It’s not quite true that 2S is a slam dunk after doing 4p. Yes I^2R and IVf loss/heat get easier, but inductor ripple is maximum at 50% duty factor, about 3x worse than for 25% duty factor, so you enter discontinuous mode earlier and either need to start pwm earlier (not so bad), or you need more inductance or higher frequency, either probably at the expense of some losses for the same footprint, but cheaper than the same game at 1S. Probably PWM earlier is as good of a solution.
Like you, I don’t see output ripple being a big issue. As I interpret the capactor dissipation factors, I’m getting values of next to nothing (1 mOhm) for ESR at 300khz. I’m inclined to stick 20mOhm in the math just in case, which still leaves output ripple under 2%, actually a tiny fraction of that at high power. The hard part here is likely good connections and traces.
Anyway, Rufus, ripple kills LED’s? and PWM doesn’t? I don’t believe it. 6A modes with 10% current ripple kills LEDs. That’s not exactly the same as “ripple kills led’s”. Of course we're talking about ripple voltage though, and that translates to more in current, still it's overvoltage/current that kills it.
As far as PWM, actually it probably doesn’t need another FET at all, just use the PWM port on the IC as it was designed. You just need one trace and a spare mcu pin, no component. Then it's up to software.
Ok, making 2S is probably somewhere between easy and impossible, I haven't looked. You should be able to cut traces (no?) and solder onto the top connections of the LED’s (after scraping the coating). Apparently that’s a pretty big pain after they are already on a heat sink. Or just sand off the trace coverlays and solder to the board. This is really too bad though. Why not have jumpers/solder pads on the LED boards? Is it really too late to do that? I doubt there can be a more capable driver with present tech, than a 4s to 2s buck. You can get melt-down modes, and suck the life out of the batteries to maintain intermediate modes, although moonlight will probably end up slightly more efficient on the 1S. so like 500 hours instead of 350.