Of course it would be. But it is good to have a backup plan, if the community will reject your changes. [...]
Also, you can try to convince the current miners, to accept a proposal, where they will send some of the current coinbase rewards to some future block numbers. Or: you can make a mining pool, which will produce such block templates, and try to encourage miners to join it.
That's maybe more interesting than I thought at first. At least there seems to be an "intermediate" group of possibilities where no hardfork and no mandatory tx fee burn is needed. It would not be a real tail emission but it could achieve a similar effect (to make rewards more predictable).
One could for example nudge people towards such "future-miner-friendly" behavior. For example creating an output type which has half of the weight units of a normal output (and thus costs only half of the fees, or any other percentage), but spends automatically a certain fee to a future block. Or a transaction type. Have to think about that a bit ...
The principle would be: : if you use this kind of output/transaction, you pay a lower fee now, but transfer a small amount of coins to future miners (best would be a fixed block number ahead).
This would make blocks slightly bigger due to this new "discount" but remain probably close to cost-neutral for users if it's designed well (depending on congestion / fee level).
The only "losers" of this model would be normal full nodes, who have to invest a little bit more into hardware, but not by much. If I'm not understanding something wrong.
But that is a problem. If blocks are empty for 4 years, then there would be no tail emission for the next 4 years.
That's true, at least there is a dependency of the past usage. However, if usage is very low, then we maybe don't need that much security.
I checked your demurrage idea and so far I cannot find any real flaw. The only problem is that this would probably never be accepted by the Bitcoin community, because above all the "hodler" fraction would consider it a confiscation. It could however be a measure of last resort -- for example, if a model like the "tail emission based on burnt transaction fees" doesn't work.