This commit replaces the initial allocation size for the content caching buffer by a `FastByteArrayOutputStream`. With this variant, allocations are cheap and we don't need to apply heuristics anymore to guess the best initial buffer size. See gh-29775