nx_wasm_vm_bounds_20260912.nx

nx_wasm_vm.nx F618 wasm-SIMD FLIP (2026-07-21, landed from the proven rung-1 twin). Adds interpretation of the 0xFD-prefixed wasm-SIMD subset the nx_wasm_v128 emitter produces (v128.load, i8x16.sub_sat_u, v128.or, i16x8.extadd_pairwise_i8x16_u/_s, i32x4.extadd_pairwise_i16x8_u, i32x4.extract_lane), byte-level-exact per the spec, so the faithdiff-class differential can prove SIMD == scalar == native. v128 VALUES ride the untyped i64 stack as POINTERS into a 4096-slot x 16B ring arena (values are consumed within an op sequence; the SAD kernel holds <=3 live). wm_skip_imm ALSO learns 0xFD -- without that the structured-control forward scanners desync on any function containing SIMD. FLIPPED into the shared VM 2026-07-21 (F618); nx_v128_gate T2/T3 + faithdiff hold it GREEN. nx_wasm_vm.nx -- a SOVEREIGN WebAssembly virtual machine: parses a .wasm BINARY and executes its exported functions over a linear memory, pure NishiLang, no third-party runtime (no node/wasmtime/wasmer). Purpose: close the last verification gap for shipped wasm -- run the COMPILED nx_vcodec.wasm offline and prove it computes bit-exact to the gate-proven NATIVE codec (a functional gate on the actual shipped bytes, which most toolchains never do; they trust the compiler). This is R0+R1: the parser (sections -> module) + a stack-machine executor for the opcode subset our nx_wat_compiler emits (i32/i64 const/arith/shift/compare, local.get/set/tee, i64/i32 load/store(+8/16), call, block/loop/if/else/end/br/br_if/br_table/return/drop/select, memory.size/grow, wrap/extend). license_tier: ORIGINAL