Deblank cuts LLM tokens 30% via AST-safe stripping

This is a pragmatic win for the LLMs are expensive era, turning human-readable code into token-efficient machine fodder without breaking the logic. A 30% reduction for Java and C++ can significantly extend context windows for legacy migrations, while the bidirectional, AST-safe nature ensures the code remains logically intact and human-reversible. With an average latency of 76ms, it is fast enough to integrate as a transparent preprocessing layer in IDEs or orchestration frameworks.