Tableau: unstable pagination, fatal RATE_EXCEEDED, and silent truncation on large Tableau Cloud instances

Describe the bug

Several defects in the Tableau connector surface on large Tableau Cloud instances (thousands of datasources, frequent catalog changes):

1. Unstable cursor pagination — paginated Metadata API queries have no orderBy, so when the catalog changes mid-ingestion, items shift between pages and entities are skipped or duplicated.
2. RATE_EXCEEDED is fatal — when Tableau Cloud throttles queries (cost budget exhausted), the RATE_EXCEEDED GraphQL error raises a fatal RuntimeError and crashes the pipeline, instead of being retried like timeouts.
3. NullValueInNonNullableField crashes ingestion — when the 20k node limit truncates a response, nested objects (e.g. workbook.owner) come back null. This error lacks severity: WARNING, so the existing non-fatal handling doesn't catch it and ingestion crashes, even though the partial data is usable.
4. Field upstream page size silently truncated — page_size * 10 (e.g. 2000 for page_size: 200) exceeds Tableau's fieldsConnection limit of 1000, which silently truncates results with no error, producing incomplete field-level lineage.

To Reproduce

Run the Tableau connector against a large Tableau Cloud instance with page_size: 200, ongoing catalog changes, and datasources large enough to hit the 20k node limit. Observe skipped/duplicated entities, pipeline crashes on throttling or null fields, and incomplete field upstreams.

Expected behavior

Ingestion should paginate stably, retry on RATE_EXCEEDED, tolerate NullValueInNonNullableField partial data, and not silently truncate field upstream results.

Version

DataHub ingestion (Tableau source), current master.