Fuse complex conversion with function application for symmetric (Juli…

…aLang#55391)

This avoids allocating an intermediate array, which reduces allocation
slightly.
```julia
julia> S = Symmetric(diagm(0=>-rand(100)));

julia> @Btime $S^0.2;
  479.196 μs (25 allocations: 560.20 KiB) # nightly v"1.12.0-DEV.994"
  478.213 μs (23 allocations: 558.58 KiB) # This PR
```

stdlib/LinearAlgebra/src/symmetric.jl

@@ -821,7 +821,7 @@ function ^(A::Symmetric{<:Real}, p::Real)
     if all(λ -> λ ≥ 0, F.values)
         return Symmetric((F.vectors * Diagonal((F.values).^p)) * F.vectors')
     else
-        return Symmetric((F.vectors * Diagonal((complex(F.values)).^p)) * F.vectors')
+        return Symmetric((F.vectors * Diagonal(complex.(F.values).^p)) * F.vectors')
     end
 end
 function ^(A::Symmetric{<:Complex}, p::Real)
@@ -853,7 +853,7 @@ function ^(A::Hermitian{T}, p::Real) where T
             return Hermitian(retmat)
         end
     else
-        return (F.vectors * Diagonal((complex(F.values).^p))) * F.vectors'
+        return (F.vectors * Diagonal((complex.(F.values).^p))) * F.vectors'
     end
 end
 
@@ -983,7 +983,7 @@ for func in (:log, :sqrt)
                 end
                 return Hermitian(retmat)
             else
-                retmat = (F.vectors * Diagonal(($func).(complex(F.values)))) * F.vectors'
+                retmat = (F.vectors * Diagonal(($func).(complex.(F.values)))) * F.vectors'
                 return retmat
             end
         end