com.atproto.sync.listReposByCollection Lexicon, for collections directory

[bnewbold]

Pairs with: bluesky-social/indigo#929

Have not run codegen in this repo; usually do that just before merge (after approval).

There was some back and forth on whether we should return array-of-objects or just array-of-strings (DIDs). I think it is good to have this more consistent with listRepos, which returns array-of-objects.

@brianolson had a minimum bound on the limit parameter (50). For consistency with other endpoints, I made the Lexicon minimum here 1.

This is just the listReposByCollection endpoint; I didn't do a Lexicon for com.atproto.sync.countReposByCollection.

[brianolson]

I think the default should be much higher, paging through millions at 50/call would be bad. The db access is super cheap, this is just about what a reasonable network blob is. 32 byte did * 1000 would only be 32kB of data so at least that. Maybe default 500 or 1000, upper limit 5000 or 10000.

[bnewbold]

I think a common case is going to be folks just iterating through hundreds of accounts, not millions, right? eg, looking for all feed generator declarations, or statusphere users. I think folks doing this bulk should configure the limit; it is nice to have it smaller for devs poking around with curl.

I'll bump to default=100 and max=5000, and add a description recommending larger.

(I don't feel super strongly about this and could bump higher as well)

[devinivy]

My inclination would be to follow sync.listRepos which has a pretty similar use case. The number of repos on a host vs. within a collection seems like they both span large and small lists.

[devinivy]

@bnewbold if you do a codegen, please also add a changeset for @atproto/api to ensure the sdk gets marked for release. I usually do this with pnpm changeset.