Well I’d think the issue has been resolved. True randomness exists when using the predictive approach.
The proof lies in Opti and Elroch continuing discussion given their past history. as if nothing ever happened. This could not have been predicted, not in a million years ! Has to be truly random by definition.
Yes, the individual samples from typical pseudorandom number generators are computationally cheap. If you want to use a lot of them, that is an advantage compared to using a natural source of randomness directly.
Page 8 of this Intel document is about "true random number generators" and draws attention to this issue! It also explains the practically important point that by seeding the random number generator at intervals, you get both high performance and high security (because the sequence of numbers between two reseeding events is too short to be cracked, and the reseeding probably makes past information useless).