There’s no reason I’m aware of that centroid mode is necessary for fragment insertion + CCD. It tends to work a lot better, because spurious/relaxable sidechain clashes are reduced, but it’s not necessary.
Go ahead and try fragment moves + ccd + energycutrotamertrialsmover in full atom mode. It ought to work. You NEED that rotamer trials, though, to fix sidechain clashes. (full packing is likely to be too expensive).
Idealization shouldn’t be necessary here – it used to be more important, but now the AtomTree stores real lengths/angles instead of looking them up. Idealization is only necessary for the most stripped-down silent files and for some abinitio things, I think.