I don't know how often Sequential employees check the forum for questions, but what I know (or think I know, mostly from the Gearspace thread) is: the output of oscillator 2 as well as the output of the filter for each voice are digitized and sent to the DSP. The amp stage, panning, mixing of the 5 voices, overdrive, effects and master volume are all digital.
The SSI2130 VCO chips each have a 5-VCA mixer on board. These are used in the Take 5 to mix the sine, saw and pulse waveforms (controlled by the shape knobs) as well as noise and sub osc level and also to control oscillator mix in the mixer section. These mixes are sent to each voice's SSI2140 VCF, and also the waveform blend of osc 2 is sent to an ADC channel for use in 2->1 FM and the mod matrix. When 2->1 FM is off, osc 2's mix level controls the VCA CVs for that oscillator's waveforms which also hit the ADC, which is why with FM off the mix level affects the amount of modulation you get in the matrix from Osc 2. When FM is on, Osc 2's VCA output is automatically maxed out when sent to the DSP (but muted going to the VCF), and the FM depth is controlled in the DSP, which mixes the Osc 2 signal into the Osc 1's digitally-generated pitch CV.
I don't know how the white and pink noise are implemented but I'm thinking they're created digitally in the DSP and fed to a DAC channel that is fed into the filters mixed with the VCO waveforms. I don't know if one of the SSI2130 on-board VCAs is used for noise level or if that's handled digitally. Noise is a mod source so it's in the DSP one way or the other, whether digitized analog noise or generated digitally.
TL;DR: VCOs, sub osc, waveform blend and oscillator mix, and VCFs are analog, everything afterward is digital. Mod matrix is digital including 2>1 FM. White/pink noise is likely digital.