I haven't went to 3.2 yet purely because I run GC via Windows at home but Linux at work and sync my athlete DB over Dropbox. There is no 3.2 package yet on Linux so I am just waiting for it rather than mucking around with 2 different versions between my machines.
From what I read, first launch of 3.2 should look at your athlete DB and convert everything to JSON. So typically it created a .bak from your device for the .fit and then generated a CPX and a JSON file for each ride.
When you open 3.2, let it do its thing and keep it open once finished. Then jump to your athlete DB dir on your machine for me and let me know what the state of that is? Does it retiain everything as .bak once it converts or does it actually trim the who data structure down to 1/2 .JSON per ride you have in there?
Purely off top of my head as a thing to try though would be to open GC 3.2, let auto convert occur. Keep GC app open and then copy/paste the now converted JSON dir structure into a new directory somewhere on your machine.... Then once that copy/paste finishes, amend your GC settings to change your athlete DB and workout lib (tools>options and bottom of the general tab those 2 boxes) to the new directory you created with the converted JSON data structure.... Close and re-open GC and see if it now points at the new location and obviously you'll see if it does a re-convert of the data structure yet again.
If you are still getting issues I will upgrade this Windows machine and just go 3.2 as there is some nice changes I want. I can re-build my GC on my Linux box at work from source and stopping being lazy waiting for a package and then I will be much more help to get it sorted out for you if needs be