1. When you hit the add-on buttons the program loads some defaults to ensure that things work sensibly. For example, if the overland filters were kept when you went to the DD3 add-on, all catalog buttons would just return empty results because the overland filters doesn't make sense in a dungeon setting. Unfortunately, this also means that the settings loaded with the map is lost. Generally, it is not adviceable to use these buttons in the middle of a map. If you wish to load symbols from a different add-on, use :ICON_CATALOG: instead, this allows you to load any catalog, not just the one provided by the defaults for whatever add-on you switch to. The easiest way to reload the correct settings is to just reload the current map, you don't need to restart the program. Simply save the map (if you have changes you wish to keep), then simply pick the map from the top of the recent file list in the File menu. This will reload the map, and reload the settings from it.
2. One way of making torn edges is to place a white polygon with some fractalized edges on top of everything. The white will cover up part of the map, appearing as a tear. Add some inside glow and perhaps some blur or similar to make it look more like a tear and not just a flat poly.

Your step 3 is the problem here. You shouldn't be selecting the same line again, it is already part of the combined path. By doing that, you are basically telling the line to combine with itself.

Once you have done 1 & 2, you are ready to use Path to Poly.

From a technical standpoint, there shouldn't be any issues, as long as you make sure that external drive gets the same drive letter every time, if that ever changes, CC3+ will give errors on startup, and won't be able to find menu files or anything.

From a performance point, loading files over USB is slower. Now, CC3+ does caching, and things like computing effects doesn't load files by itself, but for example zooming may require higher resolution files to be loaded, which will be a performance hit.

As for performance, and external drive is far faster than a USB stick (assuming said external drive is an SSD, but you get some nice external drives that almost looks like a large USB stick, but with a M2 drive inside, and these are quite fast.

Windows also have different caching modes. If you set your drive to use a memory cache, performance will be better, but it is very important that you don't just yank it out in that case, but use the "safe removal" option like you always had to do in the old days. Windows switched to uncached writes a while back (at least for USB sticks, I actually don't remember the defaults for external drives) because people always forgot the safe removal thing, and corrupted their drives because they though they could remove them once the copy dialog was done, but this does mean worse performance and should really be enabled on an external drive you actually work on.

No, not reinstall CC3+ itself, but the latest update (Currently update 27). The update fixes things not just with the core CC3+ install, but also updates files from the add-ons which is why it should always be installed after installing an update (or at the end if you install several in a row) even if you have installed it previously.

That is a simplification of how things work though. Not all tasks can be parallelized (and this holds true in new modern 64-bit software too, there are many cases where you simply cannot start calculating the next step before you know the result of the previous calculation), so even if a program uses multiple cores, there will often be times where there is a longer running task running on a single core which needs to finish before more cores can be spun up with other tasks. But that doesn't mean that it doesn't utilize multiple cores in a good way.

For a simple example, if half the workload can be hugely parallelized, but half needs to run on a single thread, it still halves the execution time compared to running on just a single core, even if the other cores seems idle a lot of the time. This can happen because the parallelizable workload get split all over the cores, so with 8 cores, the computer will chug through those parts much more quickly. Of course it would be nice if it could use all the cores all the time, but that's simply not always possible.

(Not claiming CC3+ can't be improved here, just highlighting some of the general issues with multiple cores and parallel processes)