I've used Nik's HDR Efex Pro in the past, and just moved over to Macphun's Aurora 2017. The main caveat with the (now free) HDR Efex Pro is that, while it tends to give better results for me than Lightroom, it is not long for this world (the next major Lightroom update may well cause Nik's plugins to all stop working, and Google has disbanded the dev team they acquired when they bought Nik a few years ago).
Aurora 2017 is a great little tool (dev trivia: much of the dev team of Nik left and formed MacPhun after the Google acquisition). I really hate the over-processed HDR look, and try to keep in the "realistic" range, for which Aurora has a lot of presets and then all the controls to tweak once I find an "almost right" preset. I do wish I could have a little more control over which pixel got pulled from which bracket frame, but generally it is far more likely to do the "right" thing than either Photoshop or Lightroom would, and generally produces nicer output than Nik would as well (although there are a few of my older Nik HDRs that I haven't been able to beat with Aurora). They just announced an upcoming "2018" version which will be available on Windows as well in about a month, so it now becomes an option for you.
Also, note that "DNG" is not a magical format. DNG is just a wrapper for non-lossy-compressed pixel/sensor data (sensor data includes the mosaic data and is "truly raw", but it will also store a number of other pixel formats including 16-bit FP TIFF).
Coming out of Lightroom HDR the DNG is a "16-bit floating point TIFF that can contain over 30 stops of image data" (see
https://blogs.adobe.com/photoshop/2015/12/dng-pros-cons-and-myths.html). Using floating point to describe the pixels sacrifices some precision (only 11 bits of precision) for range of colors (the aforementioned 30 stops of range). At the same time, a 32-bit non-floating-point tiff offers the same range with much higher precision, albeit at the cost of a twice-as-large image file. However, support for 32-bit fixed-point TIFF is not really widespread. 32-bit floating point TIFF doubles the 16-bit FP precision, and expands the range to an unrealistic amount. The extra range is essentially wasted, but the extra precision is very helpful if you need to "pull apart" an area of the overall-HDR "compressed tone" image (ex, a dark fog which needs to have details extracted).
Also see
http://protogtech.com/adobe-lightroom/adobe-dng-hdr-format/
Overall, I tend to do any necessary gross adjustments in Lightroom (synchronized edits across all photos in the bracket), then toning steps over in my HDR program. If I can skip the first Lightroom adjustment step and just read the raw files into Aurora I do that, obviously. The output of the HDR app is a 16-bit TIFF, which is less range but more precision than the 16-bit FP TIFF embedded in a DNG. However, since I've done most of my edits in Aurora, the extra range isn't necessary going back into Lightroom.
That said, being able to preserve the extra range from the tone mapping going back into Lightroom would be much nicer. Just not as nice as having a better HDR engine than Lightroom's built-in engine combining the brackets.