I think that for all my needs GPRS (2G) will be enogh, but anyway, maybe 5G will be better.
My smartwatch will have also WiFi 6 and bluetooth 5.1.
Your design so far just consists of throwing the latest and greatest standards into one pot in stirring.
These standards are so new, there's not really any commercially available system that you could even buy for BT5.1 as far as I know. WiFi 6 (which is an incredibly stupid marketing name), better known as IEEE802.11ax is clearly something you'd want in modern high-rate devices, but on a smart watch, the MIMO features simply can not work, due to size constraints.
So, instead of just throwing together standards and thinking that'll be good, start researching what they actually technically mean and what their benefit is.
Just like you do for the power usage of 5G vs 4G, which is an excellent question!
Btw, will it worth putting a more modern SoC/SiP ?
Well, considering you won't get IEEE802.11ax nor BT 5.1 with the SoC that you (unlike Samsung) can buy today, yes, there will be no other way.
I have heard also somewhere that with 5G the devices will be able to choose what is best connection, that is, will be able to automatically switch to 4G, 3G or even 2G if the requirements won't be too demanding.
That's generally more or less true. A modern communications device will typically be backwards compatible. That's typically necessary for coverage reasons!
So, the downer first:
"5G" doesn't mean all that much to customers. The world actually just started auctioning off the spectrum necessary to actually do "5G New Radio", so while you can get phones that have the 5G software stack (most of a modern UE is software defined radio and a lot of data handling done in software with accelerators), you can't buy much New Radio hardware so far, and you'll have to find a network operator that has such a network already running. For that to happen, you'll need to find a country where there's 5G spectrum allocations already happening.
So, while the 5G software side most definitely contains clever enhancements, the difference to you would be small.
There's LTE NB-IoT (LTE Narrowband IoT) and LTE Cat-M1, both of which are designed for machine-to-machine communications at low rates with low power consumption. Could fit your bill very well!
Your first estimate of 5MB/min was worlds off: you can get at most one GPS fix per second, and you really don't need much cleverness to see that there's at most say 16 bits of difference between consecutive GPS positions then, which leaves you with less than 1 kB per minute. That could be well-covered with much lower rates.
These standards were introduced in LTE Release 15 (I think), which "technically" is 5G, but not marketed as such, because it's not technology targeted at a consumer audience, but engineers. The chips for these standards aren't compatible with 2G, 3G, 4G or the full-speed 5G modes at all – they're doing one job, and that is delivering low-rate sensor data. Coverage, at least in Germany, seems to be pretty good if you pick the right network operator.
Same applies to you having WiFi: that eats a lot of battery, and your user doesn't really profit from it! So, having more features isn't actually good. Reduce your feature set to something your user actually wants.
Generally, your specs simply don't line up with your use case:
If you want your GPS to be running all the time, bad news, your watch battery will be empty within a few hours. GPS' power consumption dwarfs most other peripherals.
You personally probably won't be building an overly feature-rich smartwatch: That requires a lot of engineering and expensive technology that you simply don't get access to. For example, the smartwatch you've referred to supports LPDDR3 RAM – an interface that means you'll have to run many controlled impedance signal lines from the SoC to the RAM chip, which means your PCB will have very many layers. In contrast, a simpler, not-so-feature-rich device could use one of the simpler mediatek chips that integrate RAM and thus have far, far fewer external connections and hence could be much simpler routed and most importantly, without having access to >100,000$ in oscilloscopes.
Start by really writing down your use case. Then, derive technical specs from that, as detailed as the use case allows: start with data rates, and a minimal viable battery life time.
From that, derive which communication interfaces would actually benefit the use case (instead of just being "nice to have"). Then make a market-based decision on what you'll actually equip your device with.
Like it is now, you're starting backwards: You've heard of so many cool interfaces and want them all. But having a 5G radio and 802.11ax doesn't make a good smartwatch.