feat(stdlib): Crypto.* — SHA-256 + HMAC-SHA256, constant-time verify (#19)

The security-sensitive counterpart to the fast, non-cryptographic Hash.*
library: standard, test-vector-backed hashing for signed saves and message
integrity, kept in its own namespace so nobody reaches for the wrong tool.

  Crypto.sha256(s)                SHA-256 -> 64-char lowercase hex
  Crypto.hmac_sha256(key, msg)    HMAC-SHA256 -> 64-char hex
  Crypto.verify_hmac(key, msg, mac)  recompute + constant-time compare -> bool
  Crypto.hex(s)                   lowercase hex of a string's bytes
  Crypto.ct_equal(a, b)           constant-time string equality

The primitives are implemented from scratch in plain integer LLVM IR
(FIPS 180-4 / RFC 2104): no libc crypto, no data-dependent branches in the
compression rounds, so a given input hashes to the same 32 bytes on every
platform and run. Digests are returned as hex strings, not raw bytes, because
a `str` is null-terminated and a raw digest can contain a NUL. MAC checks use
a non-short-circuiting compare so timing does not leak how much of a forged tag
was correct.

Emitted on demand via g_uses_cryptort, mirroring the emit_hash prelude gate.
Scoped to the deterministic, known-answer-testable core; OS-backed
random_bytes (the one piece that can't be validated by test vectors) is left
for a follow-up.

Tested against published SHA-256 vectors (empty/"abc"/fox + 55/56/64-byte
multi-block padding) and HMAC-SHA256 vectors; wired into the self-host suite as
`crypto`. Docs: a new Crypto section with honest "what this protects / does
not" guidance, one page per method, all fences checked and in the inventory.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs/language/crypto/crypto-hex.md

@@ -0,0 +1,25 @@
+---
+id: crypto-hex
+name: Crypto.hex
+category: crypto
+kind: namespace-method
+tokens: Crypto.hex
+sig: Crypto.hex(s) -> string
+tip: Lowercase hex of a string's bytes.
+order: 4
+ns: Crypto
+member: hex
+---
+
+Encodes the bytes of <code>s</code> as a lowercase hex string — two characters per byte, so an <em>n</em>-byte input becomes a <em>2n</em>-character result. It is the same encoding <a href="crypto-sha256"><code>Crypto.sha256</code></a> already applies to a digest, exposed on its own so you can render arbitrary bytes (a key id, a small binary token) in a form that is safe to print, log, or embed in text.
+
+Parameters:
+- `s` — the string whose bytes are encoded
+
+```ludic
+program HexDump {
+  entry {
+    print(Crypto.hex("abc"))   # 616263
+  }
+}
+```