GPT-EBD2N training logs

Self-contained HTML report written by the EBD2N runtime at the end of the training run behind. Opens in any browser offline; [run tag]_[job id].json holds the same data. It records, from the training process itself: the run identity (purpose, configuration, command line, host, GPUs, PyTorch/CUDA versions, git commit [commit]); the hyper-parameters (model, optimisation, numerics, pipeline depth, freeze and LoRA recipe); the final metrics (validation accuracy per epoch, loss and token accuracy per batch, throughput, peak memory per GPU, trainable parameters); interactive curves with a table view; and, for every unit, its device, state (frozen / LoRA / full rank), parameters, optimizer steps taken and skipped. The exact-match score is computed afterwards from the saved checkpoint and appears in the job log of this record, not here. Generated by ebd2n/report.py, included with its test in the code archive.