Speed up is_homogeneous

Some measurements taking the example from the `is_homogeneous` docstring:

    R, (x, y, z) = graded_polynomial_ring(QQ, ["x", "y", "z"], [1, 2, 3])
    f = x^2+y*z
    W = [1 2 1 0; 3 4 0 1]
    S, (w, x, y, z) = graded_polynomial_ring(QQ, ["w", "x", "y", "z"], W)
    F = w^3*y^3*z^3 + w^2*x*y^2*z^2 + w*x^2*y*z + x^3

Before:

    julia> @Btime is_homogeneous(f);
      1.533 μs (62 allocations: 2.28 KiB)

    julia> @Btime is_homogeneous(F);
      3.932 μs (152 allocations: 5.53 KiB)

After:

    julia> @Btime is_homogeneous(f);
      314.970 ns (6 allocations: 288 bytes)

    julia> @Btime is_homogeneous(F);
      642.638 ns (8 allocations: 520 bytes)

src/Rings/mpoly-graded.jl

@@ -978,14 +978,20 @@ true
 function is_homogeneous(F::MPolyDecRingElem)
   D = parent(F).D
   d = parent(F).d
-  S = Set{elem_type(D)}()
+  S = nothing
+  u = zero(D)
   for c = MPolyExponentVectors(forget_decoration(F))
-    u = parent(F).D[0]
+    zero!(u.coeff)
+    # TODO: once Hecke supports zero! on FinGenAbGroupElem, switch to this:
+    #zero!(u)
     for i=1:length(c)
-      u += c[i]*d[i]
+      addmul!(u.coeff, d[i].coeff, c[i])
+      # TODO: once Hecke supports addmul! on FinGenAbGroupElem, switch to this:
+      #addmul!(u, d[i], c[i])
     end
-    push!(S, u)
-    if length(S) > 1
+    if S === nothing
+      S = u
+    elseif S != u
       return false
     end
   end