(fix): size the dataset card from real shard files, not lfs pointers

(fix): verify lfs pointers by their recorded sha256 after publishing

(fix): read published shards through shard_file in the build

(fix): read published shards from the hub when the checkout has lfs pointers

(fix): rebase onto origin before publishing

(fix): correct publish's push summary and install the lfs hooks

(update): build incremental shards so daily publishes stay small

(update): publish through git so github and huggingface share one history

(fix): enforce the commit convention in publish and workflow messages

(feat): package the dataset build as ai_tech_articles