Add specialized * for Rational and Integer

[garborg]

Piggybacking off of #35444. Seeing that PR, I realized I have code that would be helped if multiplication got the same treatment. Trying following benchmark:

function mul_spec(x::Rational, y::Integer)
    xd,yn = Base.divgcd(x.den,y)
    Base.checked_mul(x.num,yn) // xd
end
mul_spec(x::Integer, y::Rational) = mul_spec(y, x)

function mk_rats(T, n)
    r = one(T):T(1000)
    Rational.(rand(r, n), rand(r, n))
end

function bench(f, rats)
    r = 1:length(rats)
    maximum(
        f(i, f(i2, f(f(rat, i2), i))).den
        for (rat, i, i2)
        in zip(rats, r, r .+ 1)
    )
end

n = 1_000
for T in (UInt, Int, BigInt)
    rats = mk_rats(T, n)
    @info "Type" T
    @btime bench($(*), $rats)
    @btime bench($mul_spec, $rats)
end

I see the following speedups:

┌ Info: Type
└   T = UInt64
  717.044 μs (2 allocations: 64 bytes)
  480.512 μs (2 allocations: 64 bytes)
┌ Info: Type
└   T = Int64
  753.683 μs (2 allocations: 64 bytes)
  485.841 μs (2 allocations: 64 bytes)
┌ Info: Type
└   T = BigInt
  4.653 ms (145006 allocations: 2.78 MiB)
  2.555 ms (80002 allocations: 1.53 MiB)

There's still a redundant divgcd in Rational's inner constructor.