[10.0] Query pipeline architecture improvements

[roji]

This issue tracks longer-term, architectural changes to the query pipeline.

Issues completed in 9.0 are visible in #31327.

Done

• Represent query parameters with a dedicated type instead of ParameterExpression #35089

Small/localized cleanups

• Make SqlNullabilityProcessor into a real ExpressionVisitor #34696
• Rewrite SearchConditionConverter #34905
• Simplify away NonQueryExpression #34768

Planned

• Consider simplifying SQL parameter names (@__city_0 -> @city) #35113
• Stop remapping lambda bodies to deal with parameters #35038
• Stop referencing the query from ProjectionBindingExpression in the shaper #32980
• Make our SQL tree fully immutable #32927

Backlog

• Use inheritance instead of annotations for TableExpressionBase (specifically temporal tables) #33044
• Query: Re-evaluate concept of pending selector in nav expansion #20291
  • Query: remove nav expansion as a way to tackle the pending selector problem  #32957
• Consider having a relational expression representing OPENJSON, json_each and so on #32576
• Stop producing duplicate subqueries in SQL #32277
• Stop referencing SelectExpression _projectionBindings (and similar) from shaper
  • Allow standalone set operations (and ValuesExpression) without a wrapping SelectExpression #32890
  • Cosmos: Reimplement ReadItem via its own expression type again #34157
• Consider doing property/navigation binding entirely from RelationalSqlTranslatingExpressionVisitor #31309
• Consider introducing nullability on SqlExpression #33889
• DRY around getting the projection from ShapedQueryExpression #31015
• Remove SelectExpression private logic out to the translation visitors, making it a minimal, regular expression node #31049
• Query: convert identifiers to be generic SqlExpressions rather than ColumnExpressions #31326
• Query: Pay your (technical) debt on time #18923
• Share optimizations between relational and non-relational providers #34152