docs: cleaner end-to-end pipeline — processor handles resize/normalize, no manual scaling

docs: add coordinate_mode docs and RTMDet end-to-end example

feat: add coordinate_mode arg (model/image/root_relative) to forward()

fix: register sqrt_s as buffer so meta→cpu materializes it

Fix auto_map + add config_class to RTMWModel

Fix auto_map + add config_class to RTMWModel

Update README: full eval table with bolded best values

Add rtmw-l-384x288 RTMW/RTMDet HF port

Add rtmw-l-384x288 RTMW/RTMDet HF port

initial commit