You say that as if Lua 5.3 and 5.4 were better than Lua 5.2 (which LuaJIT has support for most of the new features of) or 5.1, rather than merely newer. But programming languages don't decay like your teeth.
That happens with all Lua applications, because Lua has never aimed for backward compatibility from one version to the next, so applications basically never upgrade to a new version of Lua.
New applications using LuaJIT will continue to be on Lua 5.1. And applications that do upgrade their dependencies but use LuaJIT are going to be stuck on 5.1 maybe forever, too.
Yes, and there's nothing wrong with that. It doesn't result in the same degree of fragmentation in Lua as it did in the Python 2/3 split, because it's both socially accepted and usually technically easy to write code that works in both Lua 5.1 and Lua 5.4 and everything in between.
Except they are? Lua 5.3 bitwise literals are a big improvement over the builtins bitand32, bitor.. the addition of integer type makes the language more suitable for embedded systems that don't have FPU.. Lua 5.4 added const attribute for local variables..
Those are tradeoffs, and I'm not convinced they're good ones. Suitability for processors that are so small they don't have an FPU isn't relevant to LuaJIT in any case, but you've been able to compile Lua with integer numbers since 1.0.
