Ray casting is exactly what you want to do here - you need to convert your mouse position from screen space in to world space, and then figure out if the ray that it creates intersects with one of your tiles.
When you say there is no eye position... do you mean you're just using the projection matrix to convert your world-space terrain vertices in to screen space? If so that shouldn't matter, you just use the inverse of the projection matrix to convert your two mouse points ( (x,y,0), (x,y,1) ) in to world space.
I think the Steam Box would fall under the PC umbrella.
The point of a console is that every unit contains the same basic hardware; as a developer you don't have to worry about the large variety of hardware configurations available on other platforms (unless you're developing for multiple consoles at once, which is common).
Steam boxes are just PCs running Valve's new flavour of Linux. I think that steam boxes can even be produced by third parties.
For deferred renderer it is common to do it like this - foreach light draw a bounding sphere/bounding screen-aligned quad/bounding cone (it depends on type of light and your preference), send info to shader about your light (position, color etc.) and blend it.
Exactly; this mean that you'll be only be doing lighting calculations for the pixels affected by each light.