Check for a Null GraphQL Node in the Shared Transport Rather Than per Caller #3687

Problem

GitHub's GraphQL API can return null for an item in a nodes list that it could not resolve at read time, with no entry in errors. The fix for #3392 checks for this in scripts/pr_review.py's own gh_graphql(). A query is read again, and a null that persists ends the run naming where it sat. The check does not live in the shared transport, scripts/gh_graphql.py request(), and the other scripts using that transport guard per site or not at all:

• scripts/pr_cleanup.py guards one connection with an isinstance(n, dict) test.
• scripts/handoff.py guards rows through rows_of, but node["name"] for node in row["labels"]["nodes"] is unguarded and raises TypeError on a null label node.

Reproduce

Feed handoff.py's label read a row such as {"labels": {"nodes": [None, {"name": "handoff"}]}}. It raises rather than reporting.

Done when

• The null-node check sits in one place every GraphQL caller shares, most likely gh_graphql.request() raising GraphQLError, with pr_review.py's query-only re-read kept around it.
• The per-site guards it makes redundant are removed.
• A unit test covers a null node through the shared transport.

Found by the local strict review on the #3392 fix, out of that change's file scope.