Now that it's morning I think I can do a better job of explaining.
First of all, Spanish and Portuguese went the same way, /j/ > /dʒ/ and no dialects involve any intermediaries so there is no need to assume them. This is just your basic fortition. But how does an approximant go stop, you ask?
Approximants involve the articulators getting close, but not close enough to create any kind of friction or turbulence in the airflow. It just whisks out smoothly. For some reason, they seem somewhat prone to fortition, moreso than fricatives for example. If we anthropomorphize the process, we can think of it as a case of people getting annoyed with half-assing the pronounciation of the sound. So they jump from one end of the spectrum that has the least occlusion, to the polar opposite, with a full stop of airflow. (Another common change for approximants is to simply fricativize, which is just a one-step change, and not a jump).
Such polar changes occur in many kinds of sounds. A few examples: "Peripheral" stops pairs such as p and k/q, are somewhat liable to interchange. F-sounds often move all the way to the back to h-sounds (but by way of losing their labial component and retaining only the lack of voicedness). Coronals, which are fully tongue-tip-operated sounds, are liable to glottalize, which may seem more surprising than the back-of-the-tongue-operated sounds such as c/k/q which is also common. P is less likely to glottalize for some reason.
When I devise sound changes for languages, I like to "ascribe" motivation to the sound changes. There are impatient sound changes (fortition-type things), lazy ones (lenition-type things and elimination/merger of sounds) and confused (metathesis, haplology, linear transfer of features etc.).
A fourth category would be the OCD-type, where a gap in the system gets fixed by transferring a sound for no other reason from one part of the "IPA table" to another. Often this involves changes in one of the others, but they often make less sense phonologically. The Argentinian change of intervocalic /j/ > /ʃ/, which is a bit strange, fits this bill (to make up for the fact that there is a /tʃ/). As did the Icelandic change /ɪ/, /ɛ/ > /e/ and /ʏ/, /œ/ > /ø/ which seemed to want to balance the fact that there were only three back vowel heights (/u/, /ɔ/, /a/), but four front vowel heights. This change was thought to be so disgusting that the school system eradicated it in a holy war during the 40s–60s. And yet, a similar change is cropping up again. The sound system has a will of its own.