The current code of the RSSFeed Scanner is broken in regards to handling magnet URLs.
The issue is on line 319 of org.kmallan.azureus.rssfeed.Scheduler, which is supposed to differentiate between absolute and relative http URLs and converts the relative ones to absolute.
But it assumes that is cannot be anything else, and that may have been true in 2008 when this code was practically abandoned.
solved the problem for me.
I tested it and it works both with filters and when clicking on a line in the Status view. It shows nicely in the history log too.
Sure it has certain delay before it starts the actual download, but this is how the magnet is designed to work. I did not see any issues with that, although spawning a separate thread to deal with it may be a nicer solution.
It can probably be improved further to deal with upper/lower case situations or other types of URLs, but this is the minimum required change.