Prevent inference issue of BigInt(::UInt128) due to recursion

[martinholters]

Not pretty, but should do the job. Let's see what CI says...

Replaces #30032, fixes #29923.

[martinholters]

Hm, no, this seems to be only part of the story, because this type assertion is still needed:

julia/base/int.jl

Lines 770 to 772 in e209c3d

	function div(x::UInt128, y::UInt128)
	return UInt128(div(BigInt(x), BigInt(y)))::UInt128
	end

And according to my reasoning, it should now be safe to remove it.

[vtjnash]

It's the presence of the kwarg, not the recursion, that makes it so we can't infer this.

[vtjnash]

(it turns out that it's not the only recursion point also, since parse.jl:111 was recently changed to attempt to do compile-time constant propagation at m::T = base == 10 ? div(typemax(T) - T(9), T(10)) : ..., in #29892)