No, I don't "not get it", I just think you're wrong. It's not that there is NO REPLACEMENT for this completely hypothetical Fortran code. It's that there is nobody paying anyone to WRITE the replacement code. If the code is not currently available, it does not mean that the solution does not exist. Therefore, the currently available solution in your example is to WRITE THE CODE. Which is 100% possible.
Since you asked: Why would you replace a legacy system? Lots of reasons. I guess that if one is finding themselves in the situation where they are considering replacing a legacy system, that perhaps there may be a few reasons already on their mind. We are talking about replacing legacy systems with systems that are compatible with IPv6 which to me would mean that a compelling reason to convert a legacy system in this example would be 'to regain/retain future network connectivity'. For example. Is the ability for your old legacy app to remain on the network worth your company's time? At some point it is going to be UNLESS YOU REPLACE IT with something non-legacy. Either way you are addressing the same problem, it's just a matter of when.
To stick with your example further, it's not like Fortran does something that is absolutely inimitable by any other language or platform. But who cares? Fortran can exist on a non-legacy platform, and fulfill the legacy function without legacy hardware. GNU Fortran compiler, for example, doesn't even compile machine code directly, it compiles assembly language. So one could write the replacement for the legacy Fortran system in optimized assembler if one wanted to. Just because nobody is paying for that to happen does not mean that it is not possible.
Again: "TONS of legacy systems out there that can NOT be replaced with any currently available 'solutions'" is simply not true. The solutions are there (i.e. write/build/outsource a replacement), they are just not being pursued because of a lack of business incentive. This is shortsighted, and will change, when considering IPv6.