add/mul bad codegen for uint256+

[mratsim]

Seen with @arnetheduck.

for uint128, stint properly generates add + adc (see #10)

But for uint256+, the way carry is done will loop multiple time on the low half

nim-stint/stint/private/uint_addsub.nim

Lines 17 to 29 in 9993b9d

	func `+`*(x, y: UintImpl): UintImpl {.inline.}
	# Forward declaration
	func `+=`*(x: var UintImpl, y: UintImpl) {.inline.}=
	## In-place addition for multi-precision unsigned int
	type SubTy = type x.lo
	x.lo += y.lo
	x.hi += (x.lo < y.lo).toSubtype(SubTy) + y.hi # This helps the compiler produce ADC (add with carry)
	func `+`*(x, y: UintImpl): UintImpl {.inline.}=
	# Addition for multi-precision unsigned int
	result = x
	result += y

Important for WASM codesize optimization.

[arnetheduck]

the issue here is that we lack an func addWithCarry(a, b: T): tuple[v: T, overflow: bool] operation - both for normal integers and for this library - with such an operation one can write a recursive version of the algorithm that lines up the operations and carries in a single pass instead of the current inefficient comparison+add

[mratsim]

Implementation can follow recint which also uses recursive high/low limbs: https://github.com/linbox-team/givaro/blob/v4.1.1/src/kernel/recint/ruadd.h#L170

[mratsim]

Closed by array backend refactoring