Fix recurring doc test failures with the cat API. (#21561)

This failure is due to the fact that we sort on store size, which is cached. So
it might happen that the store size that is taken into account is not the right
one, which makes the indices sorted in the wrong order. This changes the doc
example to sort on the number of docs instead.

Closes #21062

docs/reference/cat/indices.asciidoc

@@ -10,12 +10,7 @@ GET /_cat/indices/twi*?v&s=index
 --------------------------------------------------
 // CONSOLE
 // TEST[setup:huge_twitter]
-// TEST[s/^/POST _flush\n/]
 // TEST[s/^/PUT twitter2\n{"settings": {"number_of_replicas": 0}}\n/]
-// We flush very early here because the index's size is cached and we sort on
-// size below. So to get a realistic sort on size we need to flush here or else
-// the size is just whatever portion of the index is pushed out of memory
-// during test setup which isn't deterministic.
 
 Might respond with:
 
@@ -64,11 +59,11 @@ yellow open   twitter  u8FNjxh8Rfy_awN11oDKYQ   1   1       1200            0
 // TESTRESPONSE[s/\d+(\.\d+)?[tgmk]?b/\\d+(\\.\\d+)?[tgmk]?b/]
 // TESTRESPONSE[s/u8FNjxh8Rfy_awN11oDKYQ/.+/ _cat]
 
-What's my largest index by disk usage not including replicas?
+Which index has the largest number of documents?
 
 [source,js]
 --------------------------------------------------
-GET /_cat/indices?v&s=store.size:desc
+GET /_cat/indices?v&s=docs.count:desc
 --------------------------------------------------
 // CONSOLE
 // TEST[continued]