Sorry, forgot to answer your question on raw LaTeX. Embed it in an HTML comment, like so:
<!-- \begin{description} -->Anything inside of a comment will be passed through verbatim to the .tex file. This is true for both MMD3 and 2. You might consider using Replacements for common codes you need to fire off. Replacements let you add your own custom replacement tokens to the compiler. It basically just works like a real-time Project Replace, without affecting the project itself, only the compiled product.
I suggest using a Binder item named “Meta-Data” at the very top of your compile. If Scrivener discovers one of these while using any of the MMD compile formats, it will tack that on to whatever meta-data is defined in the compile pane. So the easiest way to do that is do a quick compile to MultiMarkdown plain, then copy and paste the top part of the resulting text file into your project. The other thing this solves is Scrivener automatically adding “Format:complete”, which MMD3 doesn’t use and having it can mess things up in some cases. Oh, and clear out the compile pane once you do that, to avoid duplication since the Meta-Data file will be appended. No worries about getting it back in, in the next version you’ll be able to paste raw meta-data blocks into this table and have them converted to the key-value system used in the pane.
The next version of Scrivener won’t have full MMD3 integration, but it will have a few useful additions. The meta-data pane will allow re-ordering of keys you’ve added, and there will be a checkbox to disable the “Format:complete” bit. The beta for these changes should be hitting the forum soon-ish. Still wrapping up a few loose ends, but definitely grab it once it is up as it will make MMD3 easier, and will have some interesting new features, too—along with the standard bug fixes and such.