a low-pass will not reconstruct the effects of zero-order hold. Am I wrong here?

a low pass filter won't reconstruct as the 'ideal' filter does (which also cannot be built ideally but in practice will be less steep).
If the DAC could only do one single sample frequency (say 44.1k) a steep high order analog filter low pass can be used which smoothes (removes) the steps that appear at the DAC chip's output pin.
In NOS CDP's it was done this way.
When using oversampling the steps are already smaller and thus the analog low pass can be lower order and will still 'remove' (smooth) the higher frequency steps.

The practical problem, however, is DAC's must be able to handle multiple formats AND when we are using higher sample rates we want the higher frequencies (those above the audible range) to be present as well. In theory you could use a fixed higher order low pass filter dropping off just above 20kHz for the higher bandwidths as well but this would upset those that are of the opinion they can hear upto 50kHz and defeat the purpose of higher sample rates as well.

As the 'steps' filtering in modern (many times) oversampling/interpolating DACs is handled in the digital domain the analog section can be made with a low order low-pass (may even be as low as 1st order) behind the DAC and does NOT serve as 'smoothing' or reconstruction filter in this case as this is done digital. It is intended to only to remove the HF garbage that remains (after upsampling) and could affect equipment connected to the DAC. (Purrin mentioned this already)

The fact that in multi-samplerate DAC's a low pass filter isn't used for reconstruction doesn't mean that you cannot use a low pass to 'smooth' the steps that would otherwise be there. All (earlier) CDP operate with this method.

As stated in a previous post; to 'smooth' in an all-analog way the low-pass filter would have to be variable in frequency and  be of high-order which is impractical and complicated to build to say the least.