I'm accepting this, but I've also noted down a bunch of minor issues in the code. Please fix them before landing this.

Thanks, looks generally good to me !
I've pointed out only a few minor things here and there.

• tutorial: improve type setting of shell logs (cosmetic)

complete first full draft of the tutorial integrating content from the paper

LGTM, I've only reported a minor typing issue.

Thanks, this looks great, and almost there.

It occurred to me that, if we accept that archive/ and meta/ will return nothing when ls'd, we're accepting a fundamental inconsistency for them: file entries in there exist but are not user-visible.
If we are OK with that, we can also go a bit further, and find what I think is a win-win middle ground between this task and T2694.

I'm not ~100% sure is the same issue, but the symptoms are very similar. I'm still getting daily this logspam:

You're absolutely right that this is a different concern than T2459, because even when we will fix that we will not rewrite history.
On the other hand the heuristic that the loader will us is probably related to what you want here: you'll want to filter out by default the same refs that the (new) loader will filter out, and have an option to unfilter them (mostly useful for old snapshots).
If we have that, will you still want the fine-grained filtering that you mentioned above? I'm inclined to think it would be overkill. YMMV.

(only a couple of minor things noted above, but you can fix it before pushing)

We have decided to go for a more radical approach and make archive/ and meta/ not ls-able.
Separate task and update to the design document to follow.

rebase

LGTM, I've only noted down a couple of minor issues

great stuff, thanks !

patch fuse-increase-readdirplus-bufsiz.patch is in F4141491

A few consideration on the open design questions: