Note that a single multibit sample is PCM! It may be a slightly different PCM, and it may be a very high sample rate PCM, but it is PCM and it IS QUALTIZED regardless of what he says.
It may be done with 64-bit floats which I would think basically guarantees transparency, but it is still PCM. A multi-bit sample is, by definition, PCM... So, it does covert the DSD stream to PCM - that is the ONLY way to perform DSP operations, like volume change, and fades.
People who claim to do editing on DSD without converting to PCM have invariably (tacitly) redefined PCM so as to mean "PCM with a sample rate no greater than 384 kHz."
A 1-bit (DSD) stream is, as you say, an extreme form of PCM where each sample value is either +1 or -1. If you perform any calculation on such samples, the resulting values will not be thus restricted and will therefore no longer be DSD. For example, a simple volume adjustment might produce samples with values of +0.5 and -0.5 instead. Since in DSD a "1" bit means +1 and a "0" bit means -1, those new values obviously cannot be represented. The only way of getting back to DSD is running the values through a sigma-delta modulator again. This can be done a few times without ill effects, but eventually noise and distortion will start creeping up.
Even if you're only doing simple edits, cutting and splicing, DSD is problematic. With normal PCM, it is common to apply a short cross-fade when joining two pieces in order to avoid a click. DSD, obviously, does not allow doing a cross-fade. If cross-fading is not possible (or not desirable, for whatever reason), splicing at a zero-crossing is a good way to avoid clicks. Again, DSD presents a challenge since it isn't well-defined where the zero-crossings are to begin with. To find a suitable point, you must apply a low-pass filter and look for zero-crossings in the result. However, if the final playback doesn't use the same low-pass filter as you did, the zero-crossing may end up somewhere else, and you still get a click.
For all practical purposes, DSD is a nasty format. The only place it belongs is inside DAC and ADC chips, and even there it's better to use more than two levels. 1-bit converters are a thing of the 90s.