I have in my commuting days examined scratches on train windows and the way they scatterd light, but my scratches seemed to behave less interestingly than these. Mine behaved in a more rainbow-like way (without the nice colours), possibly because they had a higher component of randomness; yours seem to be largely parallel. I never noticed any tendency for the sun to lie on an arc of light on my train windows. I think, before I could suggest anything helpful, that I would like to see some scanning electron micrographs of the scratches, but even then I would like to get the opinions of catastrophe theorists and optic physicists before trying to guess at anything myself. 

The basic principle is simple enough. One may see examples when you look at a point source of light through many embossed bathroom window panes, as long as the pattern is reasonably regular. One gets a pretty, knotted pattern. I had expected to find examples online, but could not think of the right keywords. I must see whether I can produce one myself. Other, deceptively complex, examples appear when looking at a light source reflected from an LP record, in case you happen to have one that has not yet been scrapped. I'll try to attach an image.

What it comes down to is that light from the point source (such as approximated by the sun or by a small light at night) gets refracted or reflected by a patterned array of points in some systematic way and creates an image as seen from some particular array of reception positions such as inside that train carriage or inside the bathroom. This I am sure you have worked out for yourself, because all I have said amounts to what you said.

Mind you... 

On a plain surface one expects to see the reflection or direct view of the light to appear in just one spot. However, clean scratches or grooves contain surfaces at other angles than that of the surface, and they will reflect light from the same source to the eye at other points. Given parallel scratches the various reflections form curves. If I attach the picture successfully, you will see the curve created when a light to the right of the record forming a curve on the surface.  What is not so clear, because I am a rotten cameraman and could not get my camera to focus suitably onto the image rather than onto the surface of the record, is that at one point the scratches reflect the light from the same spot as the surface; that is where you see the light lie on the curve. In the view through the train window, that is the point where the light passes through at the same angle as it passes through some of the scratches.Also, there were several other details of the behaviour of the light, with multiple reflections of the source, that I could not explain or reproduce in the photo.

As you can tell, there are differences between the notionally parallel scratches on the glass and the concentric grooves on the record, but unless you have means to produce parallel microgrooves on a reflective or transparent surface, playing with an LP disk is probably the most convenient way to explore the behaviour of such systems.

This is a very interesting question which has had me thinking for some time! Initially I thought that the curvature of the arc must be due to the scratches in the window not being parallel, but on further reflection (ha ha!) I believe that you are correct and that you are indeed seeing an arc centred on the vanishing point. I'll try to explain my thoughts without using equations, but it's quite complicated visualising 3D geometry. I've attached a modified version of your photo, which may help clarify my explanation.

The sun is distant enough that it can be considered a parallel light source. Let's say that its rays hit the window at an angle A. The scratches on the window can be imagined as long thin cylinders, and depending on how far round the cylinder a ray hits it, it will be reflected in different directions. However all the reflected rays will make the same angle A with the cylinder's axis - so you can imagine a cone of reflected rays emanating from each point on each cylindrical scratch (where the angle between the scratch and the edge of the cone is A).

Now think of the plane of the window covered in these cones all pointing in the same direction (along the lengths of the scratches) and each cone making the same angle A with the scratch that runs through its centre. Wherever your angle of view is such that you are looking directly along the edge of a cone towards its apex (such as at points X and Y in the attached image), you will see a bright spot, because that angle of view is one of the possible reflections of the direction of the sun's rays hitting that point on a particular 'cylindrical' scratch.

So we are thinking of the window as a plane covered with imaginary cones! Looking towards the vanishing point (VP), which is infinitely far away in the direction of the scratches, we will be looking effectively parallel with the scratches and so directly along the axis of a cone. In any direction that makes an angle A with that direction we will be looking along the edge of a cone, and so will see a bright point. However only directions in one half of our field of view (to the left of the VP in your photo) will actually intersect the plane of the window, and so we will see a part of a semicircle of light centred on the vanishing point. The edge of the semicircle makes an angle A with the direction of the VP, and it passes through the point where we are looking directly towards the sun - as seen in your picture.

I hope that explanation makes sense!

Here:

http://amasci.com/amateur/holo1.html

is an example on what "scratches"  can produce.

Georg