When I talk about referential transparency in the context of FP, it's honestly of no consequence to me what the history of the term in philosophy is (mind you, I still find it interesting -- I didn't know this and it was nice learning about it -- but it doesn't significantly change my understanding of the term in the context of FP).

Technical terms can be overloaded, I don't think this is either surprising or problematic. Astrophysicists classify carbon as a metal; they're not wrong, it's just a peculiarity of the field that's almost always clear from context.

In the context of FP, the term "referential transparency" means that variables can be replaced with their definitions, and vice versa, without changing the semantics of the program. The term coincides with pureness, but IMO it emphasizes something different. "Pureness" is a negative definition, it the property of there not being any side effects. "Referential transparency" to me indicates what you can do with that purity: it allows you to reason algebraically about your program.

> most languages without macros are referentially transparent

I don't think so. In most languages, you cannot syntactically replace a variable with its definition and expect the semantics of your program to remain unchanged. For instance, in Python, given the following shared prefix:

    state = 0
    def f():
        nonlocal state
        state += 1
        return state
    
    x = f()

The programs

    x + x

and

    f() + f()

do not yield the same result, which is what you'd expect if referential transparency holds.