Would it be even remotely possible to force the change in sha256 bitcoin algorith to something else that asic and fpga devices couldnt compute?
No. For one thing, ASICs and FPGAs can be constructed for any computable function
He didn't say it would have to be something a CPU could do, did he? He just said something an ASIC or FPGA couldn't do. For example, an algorithm that required 1TB of memory could not be computed by a CPU, ASIC, or FPGA. (Without additional hardware that would determine the performance and be the same for all three implementations.)
ie, anything a CPU can do, an ASIC can be constructed to do it more efficiently.
That's not true either. As a silly counter example, consider this algorithm: "Given a sequence of x86 assembly instructions, run them until they return, and tell me what's in the registers". You really think you can make an ASIC that's faster at that task?
It's just a question of whether it's worth the development cost.
Exactly. What matters is not theoretical but practical. Whether it is worth the development cost depends primarily on how big an advantage the ASIC would have over a CPU. If the algorithm were constructed such that CPUs were already nearly-optimal (for example, if it required lots of branches and lots of memory), there would be no cost justification for developing an ASIC. Instead, miners would just use lots of CPUs.
I think if we had it to do over, we'd pick a mining algorithm that requires lots of memory and lots of branches. But I don't think it's at all possible to change things now.