Painless has working lambdas now but they are incomplete, can't capture any variables from the current scope.

This adds capture support. To keep it simple nodes get a new extractVariables(Set<String>) to push their variable names. We call this on the lambda body before analysis to give us the capture info we need.

Captures work slightly differently than in java. In java they must be effectively final variables. Instead in painless they are marked read-only inside the lambda body (so you cannot assign them to a different value inside the lambda, as that would not do anything).