[FollowUp] By-Span date aggregation should always not represent null bucket

[LantaoJin]

Is your feature request related to a problem?
This is follow-up of #3528. In #3528, we convert composite aggregation to data-histogram aggregation for single bucket with bucket_nullable=false.

Actually, the performance gains comes from the range_query rewrite if optimizable.

What solution would you like?
This PR is targeting to partially revert #3550:

• Revert the aggregation conversion, remaining to use composite aggregation
• When bucket_nullable=false, set missingBucket = false
• By-Span date aggregation should always set missingBucket = false