No it isn't s small price, being able to see people's faces is a very important part of human society.
Firstly there is no "the science", science isn't a binary, fixed point. Secondly there is very little, nay no evidence that masks work in non-medical scenarios, this has been covered time and again on this thread. Thirdly whilst this is a new virus, there is nothing new about how it spreads. It does exactly what all other viruses do, it spreads. Quite frankly the world has resorted to fear and stupidity in the misguided belief that this particular virus can be "beaten". Quite why is a question best put to the politicians that have made these decisions, but here is the bottom line. The virus cannot be beaten, especially with silly bits of cloth over people's faces. The statistics of it's spread post mandation proves that.
Can I take from your post that you would agree that masks have some benefit in a medical setting?
If that's the case, then given that duration is a major factor on the risk of infection then being in a medical setting (up to 12 hour shift) whilst wearing a mask is likely to need to provide a much higher level of protection than if the same mask is used in a non medical setting (i.e. in the supermarket for 45 minutes), therefore it's not too unreasonable to assume that they could provide the same benefits in the community.
Yes you might need to be aware of good training for putting on/taking off the mask to obtain that benefit, however (and I don't know the answer) is that going to alter the risk by more than other variables (including time) which also come into play (lower number of cases, both wearing masks, better/worse ventilation, variations in proximity, etc).
The reason that testing is done in medical settings is because the risk is much higher and so it's easy to identify people who are infected and you're likely to have more cases (resulting in it being easier to measure). Conversely in a community setting the level of risk is lower and so the difference between the two is likely to be more open to a bigger variation due to 1 extra/1 less case than would be the case where you've efficiently got a larger sample size.
For instance if there's about 10% difference then the change in outcome between 10 and 11 from a sample of 100 is much bigger than between 100 and 101 from a sample of 1,000. It's why sample size is key to how reliable results are.
Such changes become more obvious the smaller the percentage change. For instance a 0.1% change wouldn't really show up until you've got a sample size of 500 or more (0.5), however you couldn't be sure that it's not just a statistical error until you've got a sample of over 2,000 (2), and probably would want to be much bigger before you could be sure that it was actually the case and you hadn't just had more cases then should be the case (i.e. is it really 2, and not one where you've just identified a second one a bit earlier than you should have done).
That variation is why there's a range of certainty within reports, to allow for the potential of there being a +n/-n within the results within the samples which skews the results one way or the other. The smaller the sample size the wider that variation.