WindowBits

Log2 size of zlib's LZ77 sliding window. Pair with a CompressionAlgorithm to select the format (raw vs zlib vs gzip); this type only carries the size.

Construction is range-checked: only Default (algorithm default) or 9..15 are representable. zlib's deflateInit2 rejects 8 entirely (Z_STREAM_ERROR), so it is excluded here. Negative / offset / out-of-range values are unrepresentable — the platform applies the algorithm-appropriate sign for CompressionAlgorithm.Raw (negate) or CompressionAlgorithm.Gzip (+16) when invoking zlib.

Why a value class instead of Int: the previous windowBits: Int parameter conflated four overlapping namespaces (0 sentinel, positive log size, negated raw form, gzip-offset form) with no compile-time prevention of invalid combinations. Callers that intend a 9-bit raw window passed windowBits = 9 and silently got a zlib-format stream that the raw decompressor couldn't read. Modeling the size separately from the format eliminates that whole class of bug.