Even if the replacement can be found "quickly", how do you ensure they don't go offline in the middle of streaming? How do you ensure "enough" copies of every chunk are distributed that it, effectively, 'doesn't matter' how many go offline [at once], it will still stream?

You don't need to _ensure_ all that stuff, in the strictest sense of the word. Network hiccups happen, nodes go down. As long as the frequency of interruptions scales inversely with the number of nodes in the network, once you're on a big enough network everything will work smoothly.