PR #9536 introduced a performance regression, supposedly caused by memsetting the buffers before each use. For reference: - https://github.com/uutils/coreutils/pull/9536#issuecomment-3634835240) - [Codspeed flamegraph](https://codspeed.io/uutils/coreutils/branches/mattsu2020%3Abasenc_bounded-memory?uri=src%2Fuu%2Fbase64%2Fbenches%2Fbase64_bench.rs%3A%3Ab64_decode_synthetic&runnerMode=Instrumentation§ionId=benchmark-comparison-section-comparison-failed)