I think darktable uses the wrong raw white level for this file:
(Only a bare minimum of modules enabled, so colours are wrong.)
A tiny reduction changes clipping detection considerably:
Two silhouette approaches, one with filmic, one with sigmoid. For both, I found the reds overwhelming, and pushed them back using the simple channel mixer preset of color calibration. I used a separate instance because I found it gave nicer colours; if memory serves me well, and I understood correctly, on the primary (chromatic adaptation) instance, the R, G and B channels are not really R, G and B (see Color calibration - colorfulness - #17 by anon41087856).
_DSF3698.RAF.xmp (13.5 KB)
_DSF3698_03.RAF.xmp (9.4 KB)