Explicitly set a confidence interval in more dense vector tests

[davidkyle]

Follow on to #145078 which missed a few tests

Closes #145180

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[benwtrent]

I mean, if we are having to do this, there is obviously something wrong with how we serialize and handle default values.

If returning and "include_defaults" is true we should return 0 confidence interval in the mapping. I am supposing that mapping merges occur with defaults included to ensure merges can occur. Are we not doing this? I wonder if that simplifies all this mess?

[pmpailis]

LGTM! Thanks for handling this to fix the failing tests!

( ++ on Ben's comment to handle this more gracefully; either here or in a follow up)

[davidkyle]

The problem is the this assertion in DocumentMapper which is firing on the old node in a mixed cluster. For this mixed cluster problem we can't use a the usual solution of a creating node feature as this means passing the FeatureService and ClusterState down into the mapper. We can't change all the BWC versions so we were left with a choice of selective muting or setting confidence_interval to the old default value making the mapping representations the same in the new cluster and old cluster. This change is the latter and unfortunately that means we have to handle all the deprecation warning which trigger not just when setting the value at index creation but every time a document is indexed.

[davidkyle]

confidence_interval is deprecated, the new default value is for it to not exist