`Nil` doesn't understand SpecialFunctions functions

[ToucheSir]

This breaks outputsize when a model uses any functions from that package. For example, FluxML/NNlib.jl#629 (comment). The easiest fix is to copy

Flux.jl/src/outputsize.jl

Lines 25 to 30 in 009d984

	for f in [:copy, :zero, :one, :oneunit,
	:+, :-, :abs, :abs2, :inv,
	:exp, :log, :log1p, :log2, :log10,
	:sqrt, :tanh, :conj]
	@eval Base.$f(::Nil) = nil
	end

for one or more functions in SpecialFunctions, but I'm wondering if there's a better way.

[mcabbott]

For the immediate problem of gelu, Flux could simply add methods to all NNlib activation functions.

I don't remember quite why we decided against missing for this, maybe because it doesn't subtype Real? But it does obey erf(missing) === missing.

[ToucheSir]

I don't remember either, but the Number subtype explanation makes sense.

[CarloLucibello]

Triage note: still an issue on Flux 0.16.11, though gelu specifically is now fine:

Flux.outputsize(Dense(3 => 4, gelu_tanh), (3, 1))   # (4, 1)
Flux.outputsize(Dense(3 => 4, gelu_erf), (3, 1))    # ERROR: MethodError: no method matching _erf(::Nil)

src/outputsize.jl still defines Nil methods from a hardcoded list of Base functions only, so anything reaching SpecialFunctions (via gelu_erf, erf, erfc, …) breaks outputsize and @autosize.