Logic for taking any sort of weird MBQL query and normalizing it into a standardized, canonical form. You can think
of this like taking any 'valid' MBQL query and rewriting it as-if it was written in perfect up-to-date MBQL in the
latest version. There are four main things done here, done as four separate steps:

#### NORMALIZING TOKENS

Converting all identifiers to lower-case, lisp-case keywords. e.g. `{"SOURCE_TABLE" 10}` becomes `{:source-table
10}`.

#### CANONICALIZING THE QUERY

Rewriting deprecated MBQL 95/98 syntax and other things that are still supported for backwards-compatibility in
canonical MBQL 2000 syntax. For example `{:breakout [:count 10]}` becomes `{:breakout [[:count [:field-id 10]]]}`.

#### WHOLE-QUERY TRANSFORMATIONS

Transformations and cleanup of the query structure as a whole to fix inconsistencies. Whereas the canonicalization
phase operates on a lower-level, transforming invidual clauses, this phase focuses on transformations that affect
multiple clauses, such as removing duplicate references to Fields if they are specified in both the `:breakout` and
`:fields` clauses.

This is not the only place that does such transformations; several pieces of QP middleware perform similar
individual transformations, such as `reconcile-breakout-and-order-by-bucketing`.

#### REMOVING EMPTY CLAUSES

Removing empty clauses like `{:aggregation nil}` or `{:breakout []}`.

Token normalization occurs first, followed by canonicalization, followed by removing empty clauses.

Predicate functions for checking whether something is a valid instance of a given MBQL clause.

Schema for validating a *normalized* MBQL query. This is also the definitive grammar for MBQL, wow!

Utilitiy functions for working with MBQL queries.

Internal implementation of the MBQL `match` and `replace` macros. Don't use these directly.