Thanks Bjorn.
There were really 2 aspects that impacted my decision. The first was whether the current framework, DoJo based, would give me high development productivity, given that I could exploit some of the existing widgets and code. The second was the future-proofing and how long it would be until ESRI moved to a new framework that would leave any DoJo apps kindof orphaned, because as you say, DoJo2 would cause breaking changes as it has a much more opinionated framework.
I got some erroneous info about DoJo2 being imminent which gave me concern about the issue 2 of future-proofing but Robert (and you) have reassured me that it's not anytime soon if at all.
Regarding issue 1, I've gone ahead with Angular2 and found Angular2 + JSAPI provides a very productive environment. The API is at a high enough level that I can quickly develop map and layer functionality. The bonus is that I'm decoupled from ESRI's implementation of the API and I can migrate to newer versions without breaking (exception: quirky name changes like map to Map) - the strategy is to avoid using visual UI elements as they have the potential to cause issues with Angular's rendering/binding framework bits. But I still have been able to load the Legend dijit into Angular and can plug it in just fine, so if there are widgets that exactly meet my needs I can probably integrate them.
Thanks for following up on this issue.