Content-agnostic voting is harmful.

Sort of. Some content agnostic voting is inevitable simply because you can't not control what people do. The way curation rewards are designed, every vote has a small positive expected return (i.e. you can't lose money by voting, ignoring RC costs, but you can make money), so indeed it is better to blindly vote than to not vote, however it is intended that blind voting earns less. That is probably not ideal but it is by design and not really a flaw or exploit.

If content agnostic voting earns a similar or higher return compared to genuine curation, then we have a problem, because it can become systemic, as it largely did under HF19/HF20 via rampant self-voting and paid voting.

Things like free downvotes (new) and the early-voting penalty (not new) are intended to mitigate that.

So in that sense (rampant content agnostic voting in HF19/HF20 plus new measures meant to reduce it), it is not at all surprising that some strategies which may have worked well under HF19/HF20 will not work as well under HF21. An obvious candidate being front running of frequent bidbot users.

I doubt those mitigating measures will solve everything, so ongoing analysis is useful.

If that really is an "important design principle", then that's a design flaw as there is no way to keep anyone from creating multiple accounts like on Voice that has a system in place that stores a unique hash of personal identification information.

It's not really a design "flaw" it is a different design. Voice would fail if people are able to bypass the personal identification mechanism, because they are counting on it working (in part because I believe they are planning to give free tokens to each account, and possibly for other reasons). Steem does not count on it, it counts on people having multiple accounts not giving them any (or much) advantage, which both reduces the incentive to do so and prevents most harms if they do.

has upvoted nearly 200,000 times since the account was created in August 2016. My account was created about six months later and has voted about 26,000 times.

Number of votes doesn't particularly matter unless it gets to the point of being a spam issue, and RC appears to have been effective in addressing the latter.

The entire premise of PoB relies on most people voting thoughtfully.

Most stake is what matters, not most people. Which is why, indeed, the failures of HF19/HF20 were so harmful. A huge amount of stake was devoted to self-voting and vote-selling.

That is why the only solution can be gently nudging those users who have a reputation to lose towards voting more responsibly

I don't know if I would say that is the only solution but it is certainly a big part of it, and a big part of that is downvotes. Other measures are possible to reduce the profitability of some forms of content agnostic voting and even derive net value from such voting. For example, the early voting penalty derives net value from bots using dumb strategies which are easy to copy competing with each other to be first on known-payout content, because a larger and larger share of the rewards are returned to the pool on content where curation is easy and rewards are not needed, essentially making the reward pool larger without increased inflation.