P0.4: wire the logic verification gate (sx assert helper + snapshot runner)

Add a real logic-test gate so future pure-sx game logic fails the build on a
bad assertion:

- tests/test.sx: `expect(cond, msg)` assert helper — prints a greppable
  `FAIL <file>:<line>: <msg>` and exits non-zero via process.exit on failure.
- tools/run_tests.sh: snapshot runner mirroring sx/tests/run_examples.sh; runs
  each tests/<name>.sx and diffs stdout + exit code against tests/expected/.
  Exits 0 iff all tests pass.
- tests/arith.sx (+ expected snapshots): seed passing sanity test.
- README.md: document both halves of the gate — logic runner and the
  reproducible ios-sim build/launch sequence (with device discovery).

tests/arith.sx

@@ -0,0 +1,13 @@
+// Trivial logic-gate sanity check: exercises the `expect` helper on passing
+// assertions and prints a stable summary line for the snapshot runner to diff.
+// Real board-state tests arrive in Phase 1.
+#import "modules/std.sx";
+t :: #import "test.sx";
+
+main :: () -> s32 {
+    t.expect(2 + 2 == 4, "two plus two is four");
+    t.expect(7 % 3 == 1, "seven mod three is one");
+    t.expect(10 - 4 == 6, "ten minus four is six");
+    print("ok: arithmetic\n");
+    return 0;
+}