Fixing reported heap corruption when releasing native memory

[drivehappy]

The Visual Studio debugger was reporting that there was heap corruption when attempting to call Marshal.FreeHGlobal. It appears that this uses a different heap than what the native side uses for allocation with malloc. To ensure that the same heap is used, a new export from zopfli.dll was added (ZopfliFree) to perform the companion free call - and the managed side now calls into this.

Since I couldn't find the original (assumed modified) sources for the zopfli.dll, I had to re-implement the ZopfliPNGExternalOptimize exported function as well to ensure the PNG compression still worked. For future reference, the implementations are:

int ZopfliPNGExternalOptimize(const unsigned char* datain, int datainsize, unsigned char** dataout) {
	ZopfliPNGOptions opts;

	const std::vector<unsigned char> origpng_cc(datain, datain + datainsize);
	std::vector<unsigned char> resultpng_cc;

	int ret = ZopfliPNGOptimize(origpng_cc, opts, false, &resultpng_cc);
	if (ret) {
		return ret;
	}

	int size = static_cast<int>(resultpng_cc.size());
	*dataout = (unsigned char*)malloc(resultpng_cc.size());
	if (!(*dataout)) {
		return ENOMEM;
	}

	memcpy(*dataout,
		reinterpret_cast<unsigned char*>(&resultpng_cc[0]),
		resultpng_cc.size());

	return size;
}

and,

void ZopfliFree(void* mem) {
	free(mem);
}

I used the Zopfli sources from the 1.0.2 release and I rebuilt the DLL binaries in Release under VS 15.7.1, and statically linked the runtime into it.