chocopan/openvla_oft_plus_int8

residual_adapter v2: retrained on all-suite stratified 960/80 (task-level holdout val base 0.05468 -> corrected 0.04490, step 4000, source chocopan/openvla_oft_residual_mlp_t4_v2)

Add trained residual MLP action head (copied from chocopan/openvla_oft_residual_mlp_t4/final, step 1500)

Add OpenVLA-OFT+ INT8 model

initial commit