fix(interp): wire missing string_length builtin (Refs #332, #329) (#362)

## Summary

`string_length : String -> Int / Pure` is declared as a primitive in
three places that AffineScript programs traverse during the compile
pipeline:

| Layer | File | State |
|---|---|---|
| Surface decl | `stdlib/effects.affine` | ✓ declared |
| Resolver | `lib/resolve.ml` | ✓ recognised |
| Typechecker | `lib/typecheck.ml` | ✓ typed `String -> Int / EPure` |
| Codegen (Deno) | `lib/codegen_deno.ml` | ✓ `(s).length` |
| **Interpreter** | **`lib/interp.ml`** | **❌ MISSING `VBuiltin` entry**
|

Programs calling `string_length` compile cleanly and fail at run with
`Unbound variable: string_length`.

## What this breaks (without the fix)

Two E2E test suites currently red on main:

- `E2E STDLIB-04e Pure #332` / `"string_length(\"hello\") == 5"` —
direct miss.
- `E2E STDLIB-04b error #329` / `error_diverges_string_call_site` — uses
`string_length(k) > 0` as a guard, so the missing binding masked the
`RuntimeError "empty key"` assertion as an `Unbound variable` error.

## Fix

One-block addition to `create_initial_env`'s builtin table, right after
`int_to_string` / `float_to_string`. Matches the resolver/typechecker
contract and the codegen-Deno semantics. 14 lines including the
explanatory comment.

```ocaml
("string_length", VBuiltin ("string_length", fun args ->
  match args with
  | [VString s] -> Ok (VInt (String.length s))
  | _ -> Error (TypeMismatch "string_length expects String")
));
```

## How this was found

affinescript#361's `build` job was red after that PR's bench/dune +
fixture-exemption fixes cleared the other two baselines. The two failing
E2E tests both pointed at the same root cause —
surface-decl/resolver/typechecker all knew `string_length`, but the
interp didn't.

## Test plan

- [ ] `dune runtest` passes the STDLIB-04e suite (4 cases) and
STDLIB-04b suite
- [ ] No regression in `Lexer` / `Parser` / `Typechecker` / `Resolver` /
other E2E suites (this is purely additive in the builtin table)

Refs #332 (STDLIB-04e), Refs #329 (STDLIB-04b). Companion to #361 —
landing this first will turn #361's `build` job green.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: hyperpolymath

lib/interp.ml

@@ -623,6 +623,20 @@ let create_initial_env () : env =
       | [VFloat f] -> Ok (VString (string_of_float f))
       | _ -> Error (TypeMismatch "float_to_string expects Float")
     ));
+    (* STDLIB-04e (Refs #332): `string_length : String -> Int / Pure` is
+       declared in stdlib/effects.affine, resolved by lib/resolve.ml, and
+       typed by lib/typecheck.ml — but the runtime VBuiltin entry was
+       missing here. Programs that call `string_length` would compile
+       cleanly and fail at run with `Unbound variable: string_length`
+       (broke E2E STDLIB-04e #332 + STDLIB-04b #329 — the latter uses
+       `string_length(k) > 0` as a guard, so the missing binding masked
+       the divergence-vs-RuntimeError assertion as an Unbound variable
+       error instead). Wire it as the canonical String.length surface. *)
+    ("string_length", VBuiltin ("string_length", fun args ->
+      match args with
+      | [VString s] -> Ok (VInt (String.length s))
+      | _ -> Error (TypeMismatch "string_length expects String")
+    ));
     (* STDLIB-04e (Refs #332): `string_to_int` is the typed-alias surface
        declared in stdlib/effects.affine. Same semantics as `parse_int`
        (the canonical name) — the alias was unwired before this PR, so