I'm an EE and I can assure you if Bitmain, Baikal, etc., truly used ASICs - rather than FPGAs - then they absolutely cannot be changed after manufacture, not even to do something as simple as a logical invert on a single bit somewhere in the algo. This is because the functionality in an ASIC is purely hardware-based, not software, and unlike an FPGA, this functionality is - quite literally - burned into it. Now, an FPGA - Field Programmable Gate Array - is usually considered a stepping stone to an ASIC in that it implements functionality in hardware, but unlike the ASIC the hardware is an array of, well, programmable gates (actually, logic function blocks connected to an I/O matrix). This allows changing the algo in an FPGA even faster than one can load a different mining program on their desktop, but the tradeoff is that FPGAs are generally an order of magnitude or more slower than an ASIC and an order of magnitude higher in cost even if every last LUT and I/O pin is used (which is never the case).

Roughly speaking, FPGAs are nearly as flexible as a GPU, use less power for a given hashrate, but can cost nearly as much; ASICs are a one trick pony with the highest electrical efficiency but which are very cheap to produce (but very expensive to develop), relatively speaking.

The change that Monero made to go to v7 is fairly minor and the current ASIC design should be easy to modify, so, yes, the next batch of ASICs could very well do both the original CN algo as well as v7, but the existing batch of ASICs can only do the original CN algo and nothing else.

Hence why I said the only surefire way to block ASICs is to change the algo - even slightly - every few months. It would still be possible to make an FPGA miner, but those are far less of a threat to decentralization than ASICs for the reasons outlined above.