Check language on globally unique ID's and update if necessary

[cholmes]

There are some changes about id's in radiantearth/stac-spec#883

We should be sure the API reflects that id's need to aim to be globally unique.

[jisantuc]

For Franklin (and, I don't know the database models, but I'm assuming arturo-stac-api as well), collection IDs and item IDs must be unique within the API instance -- this constraint is because we're using the item / collection IDs as table primary keys in Postgres. I don't know anything about what elasticsearch backed APIs require though.

[matthewhanson]

In the old sat-api there was an Elasticsearch index for collections and one for items, so Item IDs needed to be unique across everything.

With the sentinel-s2-l1a-cogs dataset we maintain a collection for both the original L2A data, and the COG converted one. I realized that I really didn't want to change the IDs of every item, I wanted the Item IDs to be the same regardless of if it was the original JP2K data or the COG version.

So in the current version of stac-server, there is an elasticsearch index for collections, and then an index exists for each collection to store those items.

[philvarner]

Should we be saying that IDs ideally are globally unique, or only that the (collectionId, itemId) tuple is globally unique?

[matthewhanson]

(collectionId, itemId) is globally unique. I don't think we should say IDs are ideally globally unique, I don't see any reason for that.