A B-tree index supports equality, ordered comparisons and ordered retrieval. PostgreSQL can choose a sequential scan when it expects to read much of a table, especially a tiny one. Creating an index doesn't promise that every query will use it.
Start from an actual workload: the columns in its filters, the order it returns and the fraction of rows it reads. A primary key already has an enforcing index. Another identical index adds write and storage costs without adding a new access path. Foreign keys don't automatically index the referencing side.
These tasks build real indexes and inspect their catalog definitions. Later plan exercises compare execution paths. The small teaching tables make correctness visible; they aren't evidence of a production speedup. Keep a query's answer unchanged while you improve access to its rows.
The referencing customer_id is the lookup key; orders.id already has a primary-key index.