Because whether purple is between blue and red on the spectrum has got nothing to do with it.

Our colour perception is mediated by the responses of our cones.

You can see the spectral sensitivity of the cones here. In passing, note that the cones don’t just respond to red, green and blue light. The L cone (long-wavelength sensitive) has sensitivity across most of the spectrum and its peak sensitivity is not even in the red region.

If we have light at, say, 440nm (as shown by the position of the grey line above) you can see that this light will activate the S cone very strongly and will activate the L and M cones only weakly. As we move along the spectrum, each wavelength of light will induce a unique ratio of cone responses. This is why we see the range of hues in the visible spectrum.

However, as soon as we consider more than one wavelength at the same time, the cone responses are no longer unique. This is why colour mixing occurs.

If we overlap single wavelength red and green light, the cone responses are identical to those that we get when we look at single wavelength yellow light. This means that our visual system cannot tell the difference between single wavelength yellow light and an appropriate combination of red and green light.

If you want to understand colour you need to think about now just observing things such as red and green light can additively mix to create yellow but also about why this is the case. That’s my opinion anyway.

Now that we have introduced these concepts we can return to the question of why an additive mixture of red and blue light doesn’t produce a hue that is between these two, say green? The reason is that the cone responses that result from an additive mixture of red and blue light are not the same as the cone responses that result from any of those wavelengths that are in the middle of the spectrum. With an additive mixture of red and blue light we get strong activation of both the L and S cones. This doesn’t happen with any of the wavelengths in the spectrum.