Wasm: fix unsigned extend to uint64 and add test

[yowl]

This widening cast was failing, particular causing a problem in

corert/src/System.Private.CoreLib/shared/System/Number.DiyFp.cs

Lines 77 to 103 in 8d81b96

	public DiyFp Multiply(in DiyFp other)
	{
	// Simply "emulates" a 128-bit multiplication
	//
	// However: the resulting number only contains 64-bits. The least
	// signficant 64-bits are only used for rounding the most significant
	// 64-bits.
	uint a = (uint)(f >> 32);
	uint b = (uint)(f);
	uint c = (uint)(other.f >> 32);
	uint d = (uint)(other.f);
	ulong ac = ((ulong)(a) * c);
	ulong bc = ((ulong)(b) * c);
	ulong ad = ((ulong)(a) * d);
	ulong bd = ((ulong)(b) * d);
	ulong tmp = (bd >> 32) + (uint)(ad) + (uint)(bc);
	// By adding (1UL << 31) to tmp, we round the final result.
	// Halfway cases will be rounded up.
	tmp += (1U << 31);
	return new DiyFp(ac + (ad >> 32) + (bc >> 32) + (tmp >> 32), e + other.e + SignificandSize);

.

Added a test that failed previously and zext instead of sext for unsigned widening. There is a parameter that I thought to use, but it is passed false here: https://github.com/dotnet/corert/blob/master/src/Common/src/TypeSystem/IL/ILImporter.cs#L574-L578. Maybe that is wrong?

Also I updated emscripten, this wasn't necessary to fix this, but I got sidetracked with a build error and upgraded. Its not a bad idea anyway as they changed the build tooling to allow for either the fastcomp or llvm backend and we will need to upgrade at some point. If you just git pull emscripten you will hit a problem with the current scripts due to these changes. Changed a bit of documentation to fit today's reality.

[jkotas]

Thanks