I'd put the diode in parallel to your coil, oriented so that a negative voltage in the coil will cause current to flow through the diode. This will clip the voltage seen at the comparator to no less than about -0.7V.

 

I have done some more experiments with my circuit. I have it (seemingly) working pretty well, but wanted to see what it was doing so I hooked it up to my sound card and recorded some more synchronised waveforms, with the coil signal on top and the output of the LM393 underneath. Consistently the circuit is dropping low at the zero crossing point which should allow me to get reliable timings. There seems to be some interference between the two signals with the tach signal being distorted when the 393 switches. That could be my sloppy circuit layout, or my method of waveform capture.