I'm extremely familiar with genetic algorithms. Pure junk is generally useless because you're no better off than starting from scratch. You need at least something that is _almost_ junk or a way to create imperfect copies of existing functional elements within a genome.
Unsurprisingly, there are mechanism for both of these. And they don't need junk DNA - bacteria can evolve just fine and they have virtually no junk DNA.
Then the question: "why junk DNA?" and the answer so far is that it has negligible fitness penalty, any observable effects become noticeable only when DNA grows to humongous size (20-30Gb) because it takes very long to replicate it.
Next question: "Then why does this plant has so little junk?". That's probably because it has some rogue transposable element that chews portions of DNA randomly.