Exact rule decoding can benefit from simple zero skips

Compared compiled exact constants, periodic sequences, counters and binary recurrences against temporal XOR plus zstd on the same wide waveform blocks. A naive rule decoder lost on CPU; skipping writes for constant-zero lanes in an already-zeroed output buffer roughly halved its decode cost without changing encoded bytes. Selecting byte rules only on size-winning blocks yielded a small storage win and a larger decode-CPU win, at higher discovery/encode cost. Charge losing candidates during selection and include layout-only controls; this experiment used deterministic rule search, not learned symbolic discovery, and establishes no whole-file ratio.