mean's dims argument is specializing as `::Function` instead of `::Colon`

[mbauman]

We use the : throughout as a sentinel for "all dimensions." We do not typically call it as a function in this use case, even though Colon <: Function.

However, I think this means that it's getting the specialization rules for Functions and thus no longer specializing as it was previously expected to. In particular, this is introducing type-instabilities that are slowing down basic functions like mean:

#### master (b72e191ea9) ####
julia> using Statistics, BenchmarkTools

julia> @benchmark mean($(zeros(10)))
BenchmarkTools.Trial:
  memory estimate:  16 bytes
  allocs estimate:  1
  --------------
  minimum time:     11.117 ns (0.00% GC)
  median time:      11.449 ns (0.00% GC)
  mean time:        13.226 ns (1.25% GC)
  maximum time:     635.868 ns (96.44% GC)
  --------------
  samples:          10000
  evals/sample:     998

#### 1.4.0 ####
julia> @benchmark mean($(zeros(10)))
BenchmarkTools.Trial:
  memory estimate:  0 bytes
  allocs estimate:  0
  --------------
  minimum time:     4.453 ns (0.00% GC)
  median time:      5.994 ns (0.00% GC)
  mean time:        5.884 ns (0.00% GC)
  maximum time:     37.309 ns (0.00% GC)
  --------------
  samples:          10000
  evals/sample:     1000

(I originally thought this was a Cthulhu bug, but it does appear that @descend is properly showing me how Julia is specializing this: JuliaDebug/Cthulhu.jl#77)

[JeffBezanson]

There used to be a method for _mean(f, A, ::Colon), and removing it probably caused this. We haven't changed the specialization rules. I think the only way to fix it is to bring back those methods, or put a static parameter on the argument.

[mbauman]

Aha, I see, this was caused by JuliaLang@97c743d