I think talking about 70-90% fee reduction in the context of CISA is a bit misleading.

Assuming a spherical blockchain in a vacuum, a P2PKH input takes 148 vB, while a Taproot input takes 57.5 vB [source]. Going from P2PKH to Taproot thus saves 1 - 57.5 / 148 = 61% of vsize on the inputs. With full aggregation you could remove the 16 vB (64 bytes) signature from each input. This would save 16 / 57.5 = 28% on the inputs compared to Taproot. While the total reduction going from P2PKH to Taproot-full-aggregation-CISA would be 1 - (57.5 - 16) / 148 = 72% per input (so lower end of your figure). Of this 72% reduction, most of it, 61% / 72% = 85%, comes from going from P2PKH to Taproot with no CISA.

Also: That's for full aggregation, which requires interactivity if there are multiple signers. There's also half aggregation, which has only half the savings, but can be performed non-interactively. If for privacy you want everyone to use the same kind of signature aggregation, then you'd have to settle on half aggregation so as not to make it impractical in certain scenarios.

I wouldn't worry too much about yet another output type being introduced after Taproot. Taproot provides privacy benefits today, such as n-of-n multisig being indistinguishable from a single-sig payment. Its privacy impact will improve as more users move from older output types to Taproot, even if a Taproot v2 comes in to split the userbase later.

There will definitely not be a hard-fork over this. Hard-forks are deeply antithetical to Bitcoin. The only way Bitcoin could have a hard-fork is if some issue puts Bitcoin in a serious crisis that could not be fixed with a soft-fork.