You forgot bid bot services. Nobody has to directly apply the vote themselves. They could even delegate to certain services, and get almost the same share of the posting rewards as if posting and voting their own non-content. There is no danger of downvotes with this method.

That's wrong. Indeed it has even been pointed out (correctly) that bid bots are more sensitive to downvotes, because they are motivated to maximize the return on their delegated vote power. Even a 20% reduction in reward means that a bid bot (or similar system) has a problem they need to address (by changing behavior such that their voting occurs with greater consensus) because their ROI is uncompetitive.

Because downvoting makes no sense with this linear rewards distribution function

No, downvoting made no sense with the non-linear rewards distribution either. The fact that downvoting consumes valuable vote power remains the same regardless of the shape of the curve. (And very few, far too few, ever engaged in it with n^2, the same as currently.)

If rewards were not linear, looking for 'popular' content (much r_shares) would actually be rewarded.
Before 'equality' hardfork, my selfvote on some shitpost by myself was 0.01 $, but it was worth 0.50* $ on a popular post

1. Curation rewards are not currently linear (they are instead sqrt-weighted). So if you are voting for the purpose of curation (as intended) any argument you are making based on linear rewards and constant 'vote value' is incorrect. Voting for popular content is currently rewarded. However, the absolute amount is too low, a problem which came into being under n^2 when the developers, with good intentions (see below), cut curation rewards from 50% to 25%, increasing the share of rewards that went to posting, and strongly incentivizing self voting and other posting-reward-harvesting schemes (which did not, at all, begin with the shift to linear, it only changed form).
2. Tacking your vote onto a popular post does not benefit you. You would not earn curation rewards; they would go to early voters. You would also not earn posting rewards unless you were engaged in some sort of self-voting or reward-sharing scheme, in which case this fails for exactly the reason I described previously (game theory devolves to stacking votes and splitting rewards or concentrating actual stake ownership into a small number of oligarchs).

you are acting like you do not understand the original intention behind the non-linear progression ?

I do understand the intention, but upon further experience and analysis, it turned out not to work. 'Original intention' does not matter one bit when the intent was based on a false (if perhaps reasonable at the time) understanding of how things work.

As we learn more about how things work or do not work, we have to move on to other ideas, not stick with 'original intention' as if it is some kind of holy book. In reality, it was really just a rapidly-thrown-together system with some good ideas and some bad ones. It is not a badge of failure that the first prototype got a few things wrong.

It was so more steem power would not result in a directly proportionally higher share of posting rewards.

Sure that was the intent. It was not only not the effect, but the actual effect was for steem power to earn more than proportionally greater rewards, both curation, and posting rewards (the latter via vote-buying, shill-posting, reward-sharing, etc.)