Deprecate fielddata on _id/_uid

[jpountz]

Some features like random_score use fielddata on _id and our documentation sometimes recommends to sort on _id in order to have a stable sort across pages (eg. https://www.elastic.co/guide/en/elasticsearch/reference/master/search-request-search-after.html). However fielddata on such unique fields has a huge cost if you have significant amounts of data, we should move away from it.

We probably have two options here, which are to either add doc values to _id (#11887) but this proved controversial due to the overhead it adds to the index, or switch to _doc but this has the drawback of not being comparable across shards or after merges.

[jpountz]

We discussed in FixitFriday and agreed to cut over random_score to use sequence numbers for random score generation. The only drawback compared to today is that documents would get a different score after they are updated, even if the same seed is used.

Consistent sorting on different point-in-time snapshots was more controversial. Here are some opinions:

• there is nothing we can do in that case, we should set expectations and just remove the recommendation to use the _uid as a tie breaker
• we should recommend using _seq_no as a tie-breaker for that purpose rather than _uid. It would not be stable in case of updates but is the best we can provide which works in practice.
• we should automatically append _seq_no to every sort internally. Same reasoning as the previous idea but without exposing _seq_no which is quite an expert/internal feature.

[rjernst]

The only drawback compared to today is that documents would get a different score after they are updated

I don't see how this is any different than the way it used to be, using the docid. This is very trappy, to have a random score with the capability of setting a seed, but the seed does not actually guarantee anything.

Why not expose the ability to set what field is used, and make it required if using a seed. Then the default can be a truly random score (using docid), but if a seed+field is set, that is used.

[jpountz]

I like that idea.