RMEP cryptographic primitive family — 37-variant performance matrix (v3, conservative measurements)

Version 1.2.0 of the RMEP performance benchmark deposit. Change from v1.1.0 (record 20218337): adopts the conservative benchmark binary (high_bit_cipher_bench.exe) as the publication source for RMC family measurements. The previous version used the more optimistic rmh_all_sizes_bench.exe numbers. Both binaries exercise the same library functions but differ in warm-up and pilot protocols, producing a 10-25% spread (e.g., RMC-2048 Chi37+QuadA R=7 at 4.68 ns/byte from the optimistic binary vs 5.82 ns/byte from the conservative binary). The full 37-variant matrix in MASTER_MATRIX.md records both. New FreeTSA RFC 3161 timestamp anchor v4 (Serial 0x04DAC6CB). Headline: RMC-2048 Chi37+QuadA R=7 = 5.82 ns/byte for 1024-bit security (4× the AES-256 ceiling); RMC-65536 = 94 ns/byte for 32 768-bit security (128× the AES-256 ceiling). RMH-256-SC Chi37 R=10 hash at 2.20 ns/byte (within 9% of SHA-3-256 software). Honest fairness disclosure: at iso-security levels that established standards can reach (≤ 256-bit), single-stream RMEP is 2-17× slower than SHA-256 (with SHA-NI), AES-256-GCM, ChaCha20-Poly1305, and SHA-3-256 on the same hardware. RMEP's contribution is (a) post-quantum security cap beyond 256-bit, (b) ~4.5× smaller R1CS constraint count for ZK circuits, (c) substrate-invariant evaluation across CPU/GPU/FPGA/photonic/quantum-atomic, and (d) single round function for hash + cipher + AEAD. Reproducibility: free-to-redistribute compiled binaries included (rmep-bench-1.0.0.tar.gz, 162 KB, SHA-256 03e7e5802708bf48a694835e0529f34e2d9d7e91af11e82d99c701b0607f0b28). Algorithm source code is proprietary IP and is not included. External anchor: 4 layered FreeTSA RFC 3161 timestamps (Serials 0x04DA4DB9, 0x04DA5884, 0x04DA5D6F, 0x04DAC6CB). Related: GitHub Release https://github.com/kiwoongeom/rmep-bench/releases/tag/v1.0.0 . Phoronix Test Suite PR https://github.com/phoronix-test-suite/test-profiles/pull/347 .