nx_sha256_wasm.nx

nx_sha256_wasm.nx -- SHA-256 (FIPS 180-4) self-contained for WAT target. Per NISHI_ANTI_AUTHORITARIAN_DOCTRINE §6, browser crypto primitives (crypto.subtle.*) are untrusted -- the substrate must ship its own SHA-256 that runs in the browser via WASM, not call out to the browser engine's implementation. This file deliberately imports NOTHING. The full nx_sha256.nx pulls in nx_syscalls (for sys_mmap) which transitively drags atomic-op IR the WAT backend does not yet lower. Here we work in caller-supplied linear-memory regions and never touch syscalls. **BLOCKED on nxc2 WAT backend codegen bug (named 2026-05-16):** The current `wasm.c` only declares `$slot_v<n>` locals for values produced by OP_ALLOCA (see wasm.c:349-355). Pointer arithmetic via OP_GEP (every `ctx_ptr[offset]` reads/writes through a GEP-derived pointer) emits OP_LOAD / OP_STORE addressing through `$slot_v<n>` without the corresponding declaration. Result: emitted WAT references undeclared locals; wat2wasm rejects with "error: undefined local variable $slot_v295". Fix path (substrate compiler work, queued): 1. WAT backend recognizes the "addr is from non-alloca SSA value" case and emits real (memory $mem) ops (i64.load/i64.store etc.) against linear memory, using the GEP-emitted offset directly. 2. Function gets `(memory (export "memory") 16)` declared so the caller (JS) can access linear memory. 3. Caller passes pointers as i32 offsets into that linear memory; substrate code does `i32.add` (current emission uses i64). 4. ALTERNATE: have WAT backend pre-scan the function for any OP_LOAD/STORE addr operand and pre-declare a `$slot_v<n>` for every such addr value. Cheaper one-line fix but allocates unnecessary locals for pointer-arithmetic intermediates. **Until that compiler work lands, the L82 chat-history hash chain uses crypto.subtle.digest in the browser (sovereignty debt named in NISHI_ANTI_AUTHORITARIAN_DOCTRINE §6 with this primitive as the queued replacement).**