[SR-15156] Another fix for Decimal(sign:exponent:significand:).

[xwu]

This issue has been present all along, and my previous fix in #3068 didn't address it because I didn't know of the problem--and (perhaps unsurprisingly) reproduced the same problem despite a totally new implementation of the initializer. Consider the following:

import Foundation

let x = Double.greatestFiniteMagnitude
Double(sign: .plus, exponent: 10, significand: x)
// +inf
Double(sign: .plus, exponent: .max, significand: x)
// +inf

let y = Decimal.greatestFiniteMagnitude
Decimal(sign: .plus, exponent: 10, significand: y)
// nan (correct, since `Decimal` has no representation of infinity)
Decimal(sign: .plus, exponent: .max, significand: y)
// runtime error (!)

This is because the implementation of Decimal(sign:exponent:significand:) converts between Int and an integer type of smaller bit width (but still plenty wide enough to represent more than twice the largest possible exponent) using a trapping initializer instead of clamping.

The present PR fixes that oversight now (in both the corelibs and overlay implementations) and adds tests.

[xwu]

@swift-ci test