That's a big mess in this thread :-(. First of all you should work in one picture of time evolution. You started to argue in the Heisenberg picture and then invoked the Schrödinger picture. These are opposite extremes of puting the time dependence either to the state kets or the self-adjoint operators that represent observables.
Here's the derivation in the
Heisenberg picture: Let ##\hat{A}(t)## be an operator that's not explicitly time dependent. Then (in the Heisenberg picture!) it obeys the equation of motion (setting ##\hbar=1##):
$$\mathrm{d}_t \hat{A}(t)=\frac{1}{\mathrm{i}} [\hat{A}(t),\hat{H}(t)]. \qquad (1)$$
On the other hand the time-evolution should be described by a unitary transformation
$$\hat{A}(t)=\hat{U}(t) \hat{A}(0) \hat{U}^{\dagger}(t). \qquad (2)$$
This implies
$$\mathrm{d}_t \hat{A}(t)=\mathrm{d}_t \hat{U}(t) \hat{A}(0) \hat{U}^{\dagger}(t) + \hat{U}(t) \hat{A}(0) \mathrm{d}_t \hat{U}^{\dagger}(t). \qquad (3)$$
Now we use (2):
$$\mathrm{d}_t \hat{A}(t)=\mathrm{d}_t \hat{U}(t) \hat{U}^{\dagger}(t) \hat{A}(t) \hat{U}(t) \hat{U}^{\dagger}(t) + \hat{U}(t) \hat{U}^{\dagger}(t) \hat{A}(t) \hat{U}(t) \mathrm{d}_t \hat{U}^{\dagger}(t) = \mathrm{d}_t \hat{U}(t) \hat{U}^{\dagger}(t) \hat{A}(t) + \hat{A}(t) \hat{U}(t) \mathrm{d}_t \hat{U}^{\dagger}(t). \qquad (4)$$
Now we have
$$\hat{U}(t) \hat{U}^{\dagger}(t)=\hat{1} \; \Rightarrow \mathrm{d}_t \hat{U}(t) \hat{U}^{\dagger}(t)=-\hat{U}(t) \mathrm{d}_t \hat{U}^{\dagger}(t). \qquad (5)$$
Using this in (4) you get
$$\mathrm{d}_t \hat{A}(t)=[\mathrm{d}_t \hat{U}(t) \hat{U}^{\dagger}(t),\hat{A}(t)]. \qquad (6)$$
From this you get by comparing with (1)
$$\mathrm{d}_t \hat{U}(t) \hat{U}^{\dagger}(t)=\mathrm{i} \hat{H}(t) \; \Rightarrow \mathrm{d}_t \hat{U}(t) =\mathrm{i} \hat{H}(t) \hat{U}(t). \qquad (7)$$
Now you can, with the initial condition ##\hat{U}(0)=\hat{1}##, rewrite this operator-differential equation as an integral equation
$$\hat{U}(t)=\hat{1} + \mathrm{i} \int_0^{t} \mathrm{d} t' \hat{H}(t') \hat{U}(t'). \qquad (7)$$
Now you can iteratively solve this equation to obtain the formal Dyson series. It is imporant to realize that in general neither the Hamiltonians at different times nor the Hamiltonian with ##\hat{U}## commutes.
The final result is the formal solution
$$\hat{U}(t)=\mathcal{T}_c \exp(\mathrm{i} \int_0^t \mathrm{d} t' \hat{H}(t'), \qquad (8)$$
where ##\mathcal{T}_c## is the time-ordering symbol. The meaning becomes clear when expanding the exponential into its power series. For the ##n^{\text{th}}## order term you get
$$\hat{U}^{(n)}(t)=\frac{\mathrm{i}^n}{n!} \int_0^{t} \mathrm{d} t_1' \cdots \int_0^{t} \mathrm{d} t_n' \mathcal{T}_c \hat{H}(t_1') \hat{H}(t_2') \cdots \hat{H}(t_n').$$
The time-ordering operator demands to order the Hamiltonians such that the time arguments are ordered from right to left in ascending order.
For a detailed derivation for this result, see
http://fias.uni-frankfurt.de/~hees/publ/lect.pdf
Sect. 1.3. Note that there I discuss a general picture of time evolution with an arbitrary initial time ##t_0##, and instead of ##\hat{U}(t)## I wrote ##\hat{A}(t,t_0)## and instead of ##\hat{H}(t)## we have an arbitrary self-adjoint operator ##\hat{X}(t)##.