I wonder if the thinking was that the object is aligned on a Single (or float) boundary - I’m not as familiar with c++ compilers as I am with c#, but I know that due to the way the memory is moved from the dimms to the cpu cache that it’s faster if the memory blocks are aligned in predictable ways and the compilers generally take care of the padding necessary to accomplish this. Given that a struct is intended to represent the memory that is the object vs a class which is intended to represent an object split across multiple locations in memory (Due to the vtable & inheritance) - it would make sense to name your structs as F as a float aligned block… This is entirely speculation though.
Not saying that Sweeney knew when he first coded it, but that would be my assumption as to the reasoning if someone were to establish F in the naming convention.