Adaptive radix tree aggregate matching

[tonyhb]

This PR introduces aggregate adaptive radix tree matching for expressions. The general idea here is to ignore expressions that don't matter given input data. For example:

• If you have 100m expressions matching on account IDs (event.data.account_id = $foo)
• And you process 10 events
• We want to evaluate 10 expressions instead of 1B expressions.

In order to do that, we must parse the CEL expression into its AST, normalize into a canonical representation, then parse each operand into an aggregate tree for matching.

Each expression is broken down into groups. For example:

a == "yes" && b == "no"

Is broken down into a single matching group containing two predicate operators for A and B. For each ident matching strings, ("a" and "b"), we create a new adaptive radix tree matching the given literal value.

When an event is received, we iterate through the keys in the given event. If there's an ART for the key in the event, we match the event's data against the tree. This returns potential expressions to evaluate.

From there, we evaluate only all matched expressions (plus any expressions that cannot be evaluated by aggregate trees). This cuts down the number of expressions evaluated by orders of magnitude.