Support more granular collation specification

Additional use cases for collation would be enabled if it could be specified at a more granular level than per-operation. Some examples:

• A compound index or sort criteria with different collations for each key
• A query filter with different collations per operator
• An aggregation pipeline with different collations per stage (e.g. a case-insensitive collation for a $match and a case-sensitive one for a subsequent $sort)
• A read-only view with different collations for the view itself, a view that it depends on, and an operation on the view (this would be enabled by allowing a collation per aggregation stage)