If Core would support to create a wallet based on BIP39 recovery words and derivation paths, it would limit such a wallet to only support the keys that such a wallet could derive.
Exactly. That's the point. If you want to create a BIP39 wallet, you can't just randomly meddle with it. It wouldn't forbid you from experimenting by adding private keys to a new wallet; it would simply allow you to choose between a BIP39 based one, or a non-BIP39 based which is what they already have.
I guess Core devs or whoever steers the development don't want such a limitation. And frankly I'm fine with that. I wouldn't want Core to be limited by above described scenario.
There's no limitation. It's just an additional feature which would only help, because people wouldn't need to download additional wallet software to connect it to their Bitcoin Core node. Someone who doesn't want all the complexities involved in Sparrow, as an example, could just download Bitcoin Core and use it simultaneously as a wallet without having to worry about losing access to their digital wallet file. One such case is someone who runs Core on a VPS and uses RPC to send and receive bitcoin.
Btw, it's perfectly possible to create a descriptor wallet in Core that completely matches a BIP39 derived wallet.
Yes, you can, in a very complex and non-human friendly way, which brings me back to the question, why don't they just adopt a mnemonic standard in the first place?
since keeping a backup of your wallet is a no-brainer, with or without a seed phrase.
Keeping a digital backup is a no-brainer to you and me, who know all the processes running behind Bitcoin Core, and who know that there's no seed phrase that can save us in case of a sudden corruption. To a random person who knows nothing, and just downloads a node because that's what he's being told, it's risky to receive any bitcoin to a wallet that does not warn the user for backups.
