Hi dc42
Your answer for using software debounce is one that I certainly would accept for large production professional designs requiring small space and small component count.
Also, as you have informed us, the inputs for the atmega device have hysterisis, then the sensor switches can be connected directly to the device, eliminating the 74HC14 completely, reducing the component count to a minimum.
However, and admitting that I do not have full details of what ironbot is doing, his project seems to me to be a learning exercise. Thus I would always advise a gate of some sort between the microproc and an outside input, to protect the microproc and allow alot of messing about with soldering, flying leads, and mistakenly connected inputs etc.
Also, as such, I would try and separate as many of the component parts of the design as possible. By using the two NAND gate debounce which is virtually infallible, we can get into the more intersting aspects of programming the functioning of the project, knowing that switch bounce is not a problem. Later on when all the fun bits have been done, then we could return to the finer points of professional design.
Incidently, as has been pointed out here already, an oscilloscope would be a very useful bit of kit to see exactly what is going on with problems like switch bounce. I've just bought a second hand scope with a terrific spec. for £100 on line.
Cheers