Regarding what I'm about to respond, a lot of it boils down to 1) me as a player, and 2) who you're making this game for.
About the encyclopedia, I am all for having something that lets you quickly look up information of things you have seen before!! I just think that the information could be more condensed, as well as not assuming that the player needs constant guidance. In a tutorial, you can also include lore or whatever else you would like to include so it flows more naturally. Usually, having stuff like this relies firstly on flavor text, and then on descriptions. While encyclopedias in real life do provide detailed explanations, unless your game focuses on the lore and on a slower pace, not many people are going to want to read that much.
On the topic of descriptions, I think that having a dual system could perhaps work. I've seen that in roguelikes like how I've described (or systems with simple-card descriptions really), there's usually stuff like that for whoever wants it. Mods for Nuclear Throne or Binding of Isaac do this, but Rounds, a game which is not really a roguelite at all, does this as well by default, with a little button in the options menu. I think this definitely could help, because the way the attributes scale up is constant. If it were different for a card, then that could be specified with an adjective, like "a little" or "by a lot", etc. And then of course, for people who want to see those stats, the option is there.
I asked a friend how long it took them to clear a movement after you asked this question, and they said they measured 2m 40s on beginner. This is still the biggest amount of time I've measured. They played some more and they said that they managed to reduce it to around a minute after using riposte. I tried to then measure my own time by playing a bunch of different rounds. The results were that for the most part, the time averaged above two minutes. At first, I kept trying to go on expert and perform mostly ripostes, but I kept getting the timing wrong and dying at almost the two minute mark without doing that much damage to the enemies. Then I decided to try to play more "casually" instead of more optimally. I managed to complete different movements, but again, more than two minutes. I reduced this time on beginner and with ripostes, but even then it still took longer than a minute and a half. The problem with this is that riposting becomes increasingly harder the more rhythms are introduced, which removes the big damage option and forces you back into the rinse and repeat playstyle, which once again averages around a little more than two minutes.
Regarding a Rhythm Heaven feel, I think that that is something that could be sort of difficult to implement in the same way, considering the tutorials are both pretty short and the levels, excluding the remixes, all focus on the singular mechanic introduced in the tutorial. One thing I think could be done is that before each movement that introduces a new enemy, a tutorial for that enemy, and that enemy only, is presented. The enemy could have way way reduced life and, similar to rhythm heaven, it could need three perfect dodges or the sort before it lets you play the actual level. This keeps a quicker and more direct flow in contrast from a sandbox the tutorial was closer to, for example. If the player then wants to practice on their own, there could be a section for that. But of course, this is all speculation.
One thing I did want to address is the thing you mention about choreographies looping. One aspect of roguelites is that since the player is going to be dying over and over, it doesn't matter if they don't get to see one thing in one playthrough, because they're sure to see it on their next, or the one after that. This is aided by slight randomization of areas, enemies and whatnot, but I'm not going to get too much into that because it's still an early demo. I just felt like mentioning this was important to take into account both for the duration and variation of movements.
This all being said of course, this is all hypotheticals and speculation that have to be proven work by playtesting, but also by what I mentioned about who you're making this for. If you want for everyone to be able to play it, then you may be better off with more than three difficulties. Crypt of the Necrodancer is definitely not for everyone (only 45.8% of players have completed zone 1 on Steam, according to the achievements, and only 4% of players have completed an All-zones mode with Cadence, the highest percentage of All-zone modes) because of its steep difficulty spike within its main story and need to be hyper-aware of your surroundings with certain characters like Aria or Bolt. And that's fine, it has still found its niche within the community (me included), but for people who just want to play a more fast-paced roguelike it probably is not what they're looking for. Your game has the ability to have insane rhythms combos, even syncopations and triplets if you really want to throw the player in for a loop, but with more complexity comes more difficulty. You can still put these elements in easier difficulties, for sure, but it's still needed to know who is going to play this.
Overall, hopefully these answers have answered the questions you had!! I'm always open to answering anything else!!