Adopt model version detection from ChatRWKV pip model python file

[schamane]

Use float for version number for model, and not a set of boolean values.

[Cyberhan123]

I prefer to use gguf logic. What I mean is that we use float for version number for model, and not a set of boolean values ​​when converting. But it is better to follow the gguf specification when loading.

[Cyberhan123]

I'm working on it

[saharNooby]

Hi! Sorry, but I don't yet see any reasons to change the existing version detection code. The current code is simple and does its job.

The proposed code uses floating-point numbers, and I'm definitely against it -- FP numbers were not designed to handle precise decimal numbers like version numbers. The proper solution is to define a version type with parsing and a comparison operator, but it would be overkill.

Moreover, there is no need to detect v6 yet, because v6 models are not supported. The code in PR would detect v6, and then create an invalid model file, which will not be readable by newer versions of rwkv.cpp when v6 support arrives (I assume v6 models will require some preprocessing, as older versions require).