perf: recalibrate vram_plan (leave GPU headroom) + refresh perf README

Fix model-intrinsic params in the model (input channels / class count / network architecture) instead of exposing them in the config

app.json: drop redundant top-level patch_size (now read from Prediction.yml)

Update model card, app.json (patch size + VRAM plan) and inference configs

Update body model

Update body model

Update body/app.json

Add ImpactSeg body

initial commit