Practice significantly complicates theory here.

A price crash that forces a bunch of miners off the network means that, not only is a 51% attack cheaper, but the spare capacity to launch one exists, and is just sitting there unused, losing money.

A volatile combination.

For security, the cost of the hashing needs to be greater than the value of being able to subvert the bitcoin network with 51% attacks.

If the cost of the hashing power is less than the value, then the incentive is to just buy the hashing power and do the 51% attacks.

So to the extent that bitcoin has any value at all, it's going to either burn an amount of electricity greater than the value of a network compromise, or it's not going to be secure.

How quickly would the difficulty adjust? My understanding is that there is an algorithm that slowly adjusts difficulty. If a big chunk of the ASIC miners went offline and mining reverted to individual GPUs, it might be a very long time before the difficulty reverted to something reasonable.

Bitcoin difficulty adjusts every 2016 blocks, nominally 2 weeks. Each time it can adjust by a factor of at most 4 (in either direction). Many other cryptocurrencies adjust difficulty every single block.