As interesting an idea as this may be, I think we need to close this as wontfix. In Ruby, a Float is a float is a float, so we obviously can't unilaterally make it a BigDecimal. In addition, using BigDecimal by default would heavily impact Float performance.
I'd be willing to look over a patch to add this feature in even a brute-force way, just to see how it looks and feels. It would require extensive changes in RubyFloat (obviously) along with changes in the interpreter and compiler where we construct literal floats and in the parser where it parses them. It's a non-trivial amount of work for something we could never have enabled by default.