Remove mention of TwicePrecision from the docs

[Seelengrab]

This type is internal and should not be mentioned in the manual, since we currently have the policy of "if it's in the manual, it's API". This is not API under semver, as its own docstring says:

help?> Base.TwicePrecision
  TwicePrecision{T}(hi::T, lo::T)
  TwicePrecision{T}((num, denom))

  A number with twice the precision of T, e.g., quad-precision if T = Float64.

  │ Warning
  │
  │  TwicePrecision is an internal type used to increase the precision of floating-point ranges, and not
  │  intended for external use. If you encounter them in real code, the most likely explanation is that you
  │  are directly accessing the fields of a range. Use the function interface instead, step(r) rather than
  │  r.step

[skip ci]

[LilithHafner]

I agree with dropping TwicePrecision from the docs, but we can keep this hint without mentioning TwicePrecision explicitly:

- In conjunction with `TwicePrecision` this can be used to implement ranges that
- are free of roundoff error.
+ When `R` and `S` are more precise than `T`, this can be used to implement ranges that 
+ are free of roundoff error.

This is how we do both Float64 ranges and Float32 ranges.

julia> typeof(1.0:2.0:12.0)
StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64}

julia> typeof(1.0f0:2.0f0:12.0f0)
StepRangeLen{Float32, Float64, Float64, Int64}

[Seelengrab]

I removed the sentence entirely because it reads as if the user is expected to do that, which isn't the case. We already do that for them, so why mention that implementation detail?

[LilithHafner]

Good point, thanks!