I'll try my best.
1. op_return was invented as a way to prevent even more destructive ways to store data. if you dig through bitcoin's history, you'll notice that in the coloerd coins / rare pepe era, people started putting data into multisig, which created unspendable UTXOs. that's just as a background. today, we're building L2s to make Bitcoin more useful as money. these need to anchor data on the blockchain and op_return is *one* way of doing that, and it's the best way of doing in in terms of minimizing damage. these L2s store legitimate data (they do not store data just for the lulz or to spam the network), but to improve bitcoin's scalability. NOTE: that this discussion is widely blown out of proportion because OP_RETURN is only economic until a pretty small payload size, but you probably know this already. nobody in their right mind would store 1 MB in an op_return if they can it in the witness, certainly not a spammer.
2. there are two important points that answer this question: fee estimation and compact block relays. fee estimation: if you don't know what the next block will look like (because you filter out transactions from your mempool), then your fee estimations will be off. if you're using an L2 that requires good fee estimation like Lightning, this even increases your risk of losing money. second, compact block filters: they minimize network data when a block is found. instead of downloading each new block from your peers, what nodes try to do is to get a summary of the block and fill in the blanks using their mempool data. in the best case, they already know all transactions and only need to download the block header etc. this reduces p2p traffic and increases block propagation speed, which in turns makes mining more competitive, and therefore more decentralized.
hope it addresses some of your quesitons.
