Approx. 60W per channel, design load is a 20-25 Ohm transducer. I went with a CT over a shunt+difference amplifier for now but either option is acceptable. I'm aiming for a 1 degree phase accuracy but if that turns out to be unrealistic I'll have to re-evaluate.
If you don't need the isolation provided by a current transformer, and at the relatively low currents involved I'd use a current shunt - much cheaper and simpler and the shunt doesn't need calibrating if you chose the right one. For example, Susumo make some low inductance SMD shunts at 1nH compared to the 3 to 5nH of typical parts:
https://www.newark.com/pdfs/techarticles/susumu/LowESLCurrentSensingChipResistor.pdfUsing a 200milliohm shunt with 1nH ESL will give a phase shift of only 0.18 degrees at 100kHz. Increasing the shunt resistance decreases the phase shift at the expense of higher dissipation in the shunt (approx 600mW for 200mohms).
It's not clear to me in your OP exactly what you need to measure. I presume you are generating a sine wave between 40kHz and 100kHz to drive a transducer, with some (unwanted) harmonics after the 11th, >= 20dBc down? I assume you don't need to measure the harmonics of the drive voltage as the filter would need to be heroic to allow (say) a 1.2MHz 12th harmonic to be within the passband but provide significant attenuation by 1.5MHz!

But what about the current waveform? If the transducer has any non-linearity the current waveform will have additional harmonics along with intermodulation products of the drive signal and its harmonics which could be significant given the latter are only 20dB down. Does this mean the current measurement has to have higher bandwidth than 100kHz? If so the phase shift becomes more of a problem, but probably still easily manageable by using a higher resistance shunt and/or paralleled to redfuce ESL.
As to the original question about calibration; the shunt shoudn't need calibration as stated earlier. The anti-alias filter will introduce several degrees of phase shift at 100kHz - but so long as the voltage and current measuring channels are using the same, reasonably close tolerance filter components the phase difference between the channels needn't exceed 0.5 degrees or so at most - simulate to confirm. Thus you can probably get away without having to calibrate the phase response.
You may need to measure both filters' amplitude response(s) to correct for droop in the passband and to ensure there are no unexpected anomolies.
[EDIT] Alternatively instead of using a C2000, you could buy an LPC-Link2 board for around $24:
https://www.nxp.com/design/microcontrollers-developer-resources/lpc-microcontroller-utilities/lpc-link2:OM13054It has an LPC4370 microcontrtoiller with a 12 bit ADC. It has a Cortex M4 and two M0 cores all running at 204MHz. Add anti-alias filters and a 2 channel MUX. The anti-alias filters can be much simpler (first or second order RC). Better still, by sampling each channel at 30 or 40MSPS you can oversample (averaging) to achieve better SNR than the C2000 ADC's.
There is free source code around for oscilloscopes using the LPC4370 including:
https://github.com/robroyx3m/High_speed_oscilloscope_using_an_NXP_LPC437xEmbedd Artists (who originally made the LPC-Link2 board) used to sell a daughter board called LabTool converting it into a 2 channel scope and logic analyser for which the schematics, firmware and PC code are available.