That is not at all what is implied by the "same as kernel modules" text in the bpf manpage. What the manpage says is, essentially, any reference to gpl_only stuff is similar to using EXPORT_SYMBOL_GPL functions in modules, that "cannot be used without the user being a derived work of kernel" according to GKH.
Now, if the userspace program that loads this eBPF is fundamentally coupled to that eBPF program, the argument carries, no?
Conversely, if your eBPF program is not a derived work of the Linux kernel (because you hand-coded its bytecode in a hex editor via reference to a clean-room eBPF spec) then loading it won't change that.
If you wrote it in C and compiled it with Clang, then whether the compiled eBPF bytecode is a derived work would require (1) looking at the source code and (2) inspecting the compiled output to see if anything unexpected got pulled in via #include.