Do not impose tol to be Float64

[lbenet]

The following throws a MethodError:

julia> using IntervalArithmetic, IntervalRootFinding

julia> f(x) = -4*x^3 + 5*x^4
f (generic function with 1 method)

julia> dom = big(0) .. big(1)
[0, 1]₂₅₆

julia> roots(f, dom, Newton, 1.0e-5)  # works ok
2-element Array{Root{Interval{BigFloat}},1}:
 Root([0.799999, 0.800001]₂₅₆, :unique)
 Root([0, 6.67704e-06]₂₅₆, :unknown)   

julia> roots(f, dom, Newton, parse(BigFloat, "1.0e-5"))
ERROR: MethodError: no method matching roots(::typeof(f), ::Interval{BigFloat}, ::Type{Newton}, ::BigFloat)
Closest candidates are:
  roots(::Any, ::Interval{T}, ::Union{Type{Krawczyk}, Type{Newton}}) where T at /Users/benet/.julia/dev/IntervalRootFinding/src/roots.jl:153
  roots(::Any, ::Interval{T}, ::Union{Type{Krawczyk}, Type{Newton}}, ::Float64) where T at /Users/benet/.julia/dev/IntervalRootFinding/src/roots.jl:153
  roots(::F<:Function, ::F<:Function, ::Any) where F<:Function at /Users/benet/.julia/dev/IntervalRootFinding/src/roots.jl:226
  ...
Stacktrace:
 [1] top-level scope at none:0

I think it makes sense to allow tol in the roots methods to be <:AbstractFloat?

[dpsanders]

Sure. I think it can be any real number in fact.

[Kolaru]

Currently we are requiring all arguments to have the same type, but this type can be anything.

This still fails, because some default arguments are Float64, so if not everything is specified, it fails.