You do in fact need an extremely steep filter. 44.1kHz is a little over an octave above 20k, and for adequate filtering and reconstruction you need 96dB of roll-off at at 16-bits and 144dB at 24-bits.
It's practically impossible to design an artefact-free filter with a roll-off as steep as that. Every single person who says that 44.1k is enough "because Nyquist" has failed to understand this.
You can trade off delay against various artefacts, including passband ripple, non-linear phase smearing, and others. But the shorter the delay, the less true it is that you get out exactly what you put in.
It's practically impossible to design an artefact-free filter with a roll-off as steep as that. Every single person who says that 44.1k is enough "because Nyquist" has failed to understand this.
You can trade off delay against various artefacts, including passband ripple, non-linear phase smearing, and others. But the shorter the delay, the less true it is that you get out exactly what you put in.