As long as your CD player runs the discrete sample values through a proper low-pass "reconstruction filter" before they're sent out its analog output, you end up with a nice, smooth waveform which looks just the way you'd want it.
First-generation CD players tended to use a "brick wall" reconstruction filter implemented in the analog domain (e.g. several stages of op amp low-pass filter). Newer ones tend to use one or another "oversampling" design, which do the first part of the filtering digitally (reconstructing numerous intermediate points between the incoming samples via a digital sinc filter) and then use a single-stage low-pass analog filter.
Now, if you were to ignore the essential requirement for a reconstruction filter during playback - if you were to just take each of these sample points, convert them to an analog voltage when it arrived, and then "hold" this analog value until the next point arrived - you would indeed end up with a rather nasty-looking waveform. It would be "stair-stepped". In the frequency domain, it would have a string signal component at 19 kHz, and numerous "aliases" of this signal at higher frequencies.
The reconstruction filter eliminates the aliases, and leaves you with just the 19 kHz signal you want.
The issue you've encountered is one of the subtler ones in CD and sampled-audio design. It's very counterintuitive to many that you actually *can* accurately sample and reconstruct a signal, using just a bit more than two points per sine... but it actually does work out that way, both mathematically and in practice (when properly implemented).