Memory footprint of stream_code during lazy transformations

When processing large datasets using the stream_code module, the goal is to maintain a constant memory footprint by avoiding loading the entire set into the heap. While the module allows for lazy evaluation, the behavior of the garbage collector and process heap growth during long-running stream transformations can be unpredictable.

Standard tail-call optimization prevents stack overflows in recursive functions, but it is unclear how stream_code manages memory pressure when streams are piped through multiple transformation stages. If the consumer process is slower than the producer, there is a risk of accumulated data chunks leading to Out of Memory (OOM) errors.

How does stream_code handle internal buffering when multiple transformation functions are applied to a large stream? Are there specific patterns to ensure the garbage collector reclaims memory from processed chunks immediately while the stream is still active?