As far as image handling goes, the best way to see what Scrivener is doing is to compile to plain “MultiMarkdown” and look at the .md file in a text editor. At the very bottom you’ll see how it is marking each image’s size.
This may or may not work for you, depending on the types of images you have, but a neat trick specific to MultiMarkdown is that if you give no image size data in this part of the file, then its LaTeX output will use this output for the graphic:
\includegraphics[keepaspectratio,width=\textwidth,height=0.75\textheight]{file.jpg}
That will work no matter the paper size, though as I say it does depend on the image type. I can’t often use this trick because I sometimes use smaller images and this blows them way up to visible blurry pixels. I suppose you could get fancy with this and make a proper script that not only checks the image line, but determines whether its specified point width is too wide for the intended output, and strip it, but otherwise leave it alone if it is smaller.
As for how to get these stripped out straight from Scrivener, you’d need to dip into the Processing compile format tab to some extent. You could either use a sed, perl or ruby one-liner here, or do that in a script and use the command here to call on the script. I’d go with a one-liner, as stripping those out would be a pretty simple regex operation.
This post covers the basics of how you can do that in the processing pane.