Add lvl3param case to TwistFFTAdd for __uint128_t support

TwistFFTAdd was introduced in commit 2b15dde to fuse backward FFT with
accumulation in CMUX, but the lvl3param branch (used by Double Decomposition
with __uint128_t) was not included. This caused a static_assert failure
when compiling gatebootstrappingtlwe2tlwedoubledecomposition.

Mirrors TwistFFT's lvl3param handling: compute via a temp uint64_t array
using fftplvl3, then accumulate (+=) into the 128-bit result.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: nindanaoto

include/mulfft.hpp

@@ -120,6 +120,12 @@ inline void TwistFFTAdd(Polynomial<P> &res, PolynomialInFD<P> &a)
         else if constexpr (std::is_same_v<typename P::T, uint64_t>)
             fftplvl1.execute_direct_torus64_add(res.data(), a.data());
     }
+    else if constexpr (std::is_same_v<P, lvl3param>) {
+        alignas(64) std::array<uint64_t, P::n> temp;
+        fftplvl3.execute_direct_torus64(temp.data(), a.data());
+        for (int i = 0; i < P::n; i++)
+            res[i] += static_cast<__uint128_t>(temp[i]);
+    }
     else if constexpr (std::is_same_v<typename P::T, uint64_t>)
         fftplvl2.execute_direct_torus64_add(res.data(), a.data());
     else