[ValueTracking] computeKnownBitsFromOperator - should we move target intrinsic handling into target hooks?

computeKnownBitsFromOperator currently handles a number of x86/riscv intrinsics and I'm intending to add handling for more x86 intrinsics in the coming months.

Before this work begins, should we instead move to a TTI hook, similar to what we do with X86TTIImpl::simplifyDemandedUseBitsIntrinsic etc?

At the moment x86 intrinsic constant folding is spread out across computeKnownBitsFromOperator / simplifyDemandedUseBitsIntrinsic / instCombineIntrinsic :(

CC @nikic @dtcxzyw

Start by reading computeKnownBitsFromOperator and compare its intrinsic handling with X86TTIImpl::simplifyDemandedUseBitsIntrinsic and instCombineIntrinsic. Trace how the existing x86 and RISC-V cases are distributed, then determine whether a TTI hook is an appropriate home; done requires an agreed design before implementation begins.

Written by the indexing model from the issue text.