Fix signed integer overflow in MakeKernelParams8bit

Compute `prod_zp_depth` via `uint32_t` casts so the intended wraparound is well-defined rather than UB. Overflow checks on the surrounding offset/pointer computations are left untouched.

See chromium issue: https://issues.chromium.org/issues/526978330 and https://issues.chromium.org/issues/520870343

Closes https://github.com/google/ruy/pull/373

COPYBARA_INTEGRATE_REVIEW=https://github.com/google/ruy/pull/373 from wangw-1991:fix_integer_overflow 781542ce3bf209d415e19d73a65e89bf073f3617
PiperOrigin-RevId: 952143874

ruy/kernel_common.h

diff --git a/ruy/kernel_common.h b/ruy/kernel_common.h
index 69e819b..d4cf293 100644
--- a/ruy/kernel_common.h
+++ b/ruy/kernel_common.h
@@ -177,7 +177,13 @@
   params->rhs_zero_point = rhs.zero_point;
   params->dst_zero_point = dst->zero_point;
   params->depth = depth;
-  params->prod_zp_depth = lhs.zero_point * rhs.zero_point * depth;
+  // prod_zp_depth intentionally relies on two's-complement wraparound to match
+  // the wrapping arithmetic performed by the asm kernels. Compute it in unsigned
+  // so the overflow is well-defined.
+  params->prod_zp_depth = static_cast<std::int32_t>(
+      static_cast<std::uint32_t>(lhs.zero_point) *
+      static_cast<std::uint32_t>(rhs.zero_point) *
+      static_cast<std::uint32_t>(depth));
   params->flags |= RUY_ASM_FLAG_NEEDS_LEFT_SHIFT;
   if (mul_params.multiplier_fixedpoint_perchannel()) {
     // Temporary release-assert to debug some crashes in an application.