Fix incremental fetch cache key generation so request bodies are represented byte-exactly rather than through lossy text decoding. Distinct bodies must not collide: account for raw bytes, body kind and fetch-relevant metadata, and use unambiguous framing for structured bodies such as FormData and files even when user-controlled data resembles delimiters. Support the standard binary and structured body forms accepted by the existing cache-key API, including ArrayBuffer and arbitrary ArrayBuffer views, Blob/File, FormData, URLSearchParams, and strings. Identical inputs must remain deterministic; views over the same byte sequence should agree, while a string and its encoded byte array remain distinct. Keep existing cache consumers working.