Well you have a number of issues in how you setup your simulation that is causing the problems.
Vertical source appearing to go horizontal
1. The horizontal axis is from 0 to 5 not -2.5 to 2.5 which means when you set the center of the source to be located at 0 um you are setting it to be right on top of the PML and even with 20 PML layers it is not enough to absorb that so what you are seeing is reflections off the edge of your domain. Shift your center over to 2.5 um and you will see what you are expecting.
Power not equaling zero
1. You are using a pulsed input which means many frequency components, the power will be spread across these. To show what you want to show you would need to add up the power for the observation lines for each frequency component. Try a CW simulation instead.
2. You are using PML as your boundary conditions. PML is an absorbing boundary condition, therefore introducing loss that you are not considering.
3. Be cautious, you are specifying the amplitude of your input plane to be 1 but want to compare power. Try specifying power so that you are comparing like quantities.
4. The initialization of the input plane is based on the paraxial approximation, i.e. field profile is much larger than wavelength. With a wavelength of 10 um and a Gaussian Full width at 1/e^2 of 1 um you are very far from this.