Score an extraction agent's amounts within a tolerance

An agent reads invoices and pulls out every amount it finds. The eval compares its amounts with a person's gold list for the same invoice, and a rounding slip of a few cents should still count as a hit.

Write match_amounts(docs, tolerance) that returns [matched, extra, missed].

• docs: a list of invoices, each [predicted, gold]: the agent's amounts and the person's amounts, as lists of whole cents in any order. Amounts can be negative (credits), and either list can repeat an amount or be empty.
• tolerance: whole cents, 0 or more.
• Return the counts, summed over all invoices: matched is the number of matched pairs, extra the predicted amounts left unmatched, and missed the gold amounts left unmatched.

Rules:

1. A predicted amount and a gold amount from the same invoice may be matched when they differ by at most tolerance.
2. Each amount, predicted or gold, is in at most one pair.
3. In each invoice, match as many pairs as possible.