It's a common misconception that pathogens "don't want to kill their hosts". In reality, pathogens don't "care" at all about killing their hosts. All pathogens "want" to do is transmit from one host to the next. (Taking out the teleogenic language: There's no natural selection on a pathogen to reduce virulence as such. The selection on a pathogen is for transmission.)
In some cases, transmission is optimized when the virus is less virulent. In other cases, transmission works just fine when the virus is highly virulent. Rabies is one of the latter cases.
If a pathogen is spread by, say, respiratory contact, then having a host that is ambulatory and seemingly healthy may allow the first host to contact many others, so reducing virulence may enhance transmission in that specific case.
If the pathogen is spread by sand fleas (as in myxoma in Australian rabbits), then rapidly killing the host is counterproductive, because fleas leave a dead body. But keeping the host healthy is also counterproductive because a healthy rabbit scratches away its fleas. Optimal transmission turned out to work with a virus that kills 50-75% of its victims, because that makes the host very sick for a long time, allowing the fleas to bite it unimpeded.
Rabies spreads by damaging the brain, making the host aggressive, and spreading through bites in saliva (among other ways, but that's an important one). Serious brain damage to the point of wildly abnormal behavior is really not compatible with a host living a long time, so optimal rabies transmission means very high host mortality.
Finally, humans are not natural hosts for rabies and rarely transmit the virus, so in any case there's minimal natural selection for any virulence changes in humans. But it likely wouldn't matter, for the reasons above.