Steadier data service; sample rows unstuck

Changes since 2.5.0.

#Fixed

• No more restarts on large files. Reading a publisher's file for a dataset preview used to load the whole file into memory before checking its size, so a few large files at once could restart the data service, interrupting OrionAI answers, search and data loading. Files are now read only up to their limit, every page of a feature service has a size limit, and only a few large reads run at a time.
• Previews of large datasets. When a dataset's file is too large to preview, the preview now reads the publisher's feature service instead, in smaller pages with simplified shapes, so boundary datasets that came back empty now show on the map.
• Sample rows for every dataset. One source value with a hidden null character stopped the hourly job that adds sample rows to dataset pages, every hour, for the whole batch. Those characters are now removed before saving, and a record that still can't be saved is set aside on its own instead of holding up the rest.