Using a tree data structure to model query pipeline #1

Currently, we represent the query pipeline using the over attribute on each SQLNode. That is, From(person) >> Select(...) results in a new Select node, Select(..., over = From(person)). While we can still compose queries by rebasing nodes recursively, having access to only the intermediate parent is a bit limiting.

We could instead use a data structure like an AST to represent the full query. Most of the annotate/resolve/link code already specializes on each SQLNode type, we only need to cast it to the tree visitor pattern. This should help fix some issues from the Julia library which are definitely present in the python port.

1. bound references are not rebased MechanicalRabbit/FunSQL.jl#13 - visit all children and update query handles after rebasing
2. do not collapse a Group node when an aggregate is used in Bind MechanicalRabbit/FunSQL.jl#12 - visit children nodes, check for a Bind node. Not collapse the Group node if present.