test(backend): cover the request-body size cap

fix(backend): reject oversized request bodies before they are buffered

refactor(backend): use Starlette's RFC 9110 status names in routes

test(backend): cover bounded upload read and at-limit acceptance

fix(backend): bound upload reads in /v1/captions

feat(backend): pull weights from HuggingFace Hub at startup when BACKEND_WEIGHTS_HUB_REPO is set

test(backend): add FastAPI route tests with fake predictor service

feat(api): build production-grade FastAPI inference backend