Grover's algorithm outputting random/incorrect results

As @narip already mentioned in the comments, the statevector simulator of the IQX (your top picture) shows that one state has 100% measure probability since you added measurements and thus the state collapses. You should only add measurements for shot-based readouts, not if you do statevector simulations.

Regarding your question about the Hadamard gate: I think there are actually some Hadamards missing! Based on your circuit I assume the oracle/boolean function you want to implement is $f(x_1, x_2) = x_1 text{ and } x_2$. The Toffoli gate with surrounding X gates you implemented indeed flips a target qubit if both qubit 1 and 2 are 0. But keep in mind that Grover's oracle must do phaseflip and not a bitflip! To convert your oracle you should add two Hadamards around the target qubit, to be

        ┌───┐     ┌───┐
   x_1: ┤ X ├──■──┤ X ├
        ├───┤  │  ├───┤
   x_2: ┤ X ├──■──┤ X ├
        ├───┤┌─┴─┐├───┤
target: ┤ H ├┤ X ├┤ H ├
        └───┘└───┘└───┘

And on top of that, you should have an initial layer of Hadamards, to initialize in an equal superposition. In total your circuit would be something like

        ┌───┐┌───┐     ┌───┐┌───┐┌───┐     ┌───┐┌───┐
   x_1: ┤ H ├┤ X ├──■──┤ X ├┤ H ├┤ X ├──■──┤ X ├┤ H ├────
        ├───┤├───┤  │  ├───┤├───┤├───┤  │  ├───┤├───┤
   x_2: ┤ H ├┤ X ├──■──┤ X ├┤ H ├┤ X ├──■──┤ X ├┤ H ├────
        ├───┤├───┤┌─┴─┐├───┤└───┘└───┘     └───┘└───┘
target: ┤ H ├┤ H ├┤ X ├┤ H ├─────────────────────────────          
        └───┘└───┘└───┘└───┘