`rationalize(x::Rational)` inconsistencies

[lvlte]

julia> x = float(pi)
3.141592653589793

julia> r = rationalize(x)
165707065//52746197

julia> rationalize(r, tol=0.1)               # tol is ignored
165707065//52746197

julia> rationalize(Int, r, tol=0.1)         
ERROR: OverflowError: 7468474 * 36028797018963968 overflowed for type Int64
Stacktrace:
 [1] throw_overflowerr_binaryop(op::Symbol, x::Int64, y::Int64)
   @ Base.Checked .\checked.jl:163
 [2] checked_mul
   @ .\checked.jl:297 [inlined]
 [3] +
   @ .\rational.jl:362 [inlined]
 [4] rationalize(::Type{Int64}, x::Rational{Int64}, tol::Float64)
   @ Base .\rational.jl:262
 [5] #rationalize#247
   @ .\rational.jl:277 [inlined]
 [6] top-level scope
   @ REPL[64]:1

julia> rationalize(Int, r, tol=0.0)
165707065//52746197

julia> rationalize(Int, big(r), tol=0.1)
16//5

We shouldn't need big(r) to reduce the precision of r.

I noticed rationalize has a single method for x::Union{AbstractFloat, Rational}. When x is a rational, the error term and tolerance term of the convergents of r (and other variables used in the function) are also rationals, and as those terms approach zero, their denominator are very likely to overflow even with a relatively large tolerance. Actually the function was written for floats (they can represent smaller values with the same precision, whereas rational components would need more). Support for rational were added in #43427.