[timlichtenberg]

Summary

The GraphQL API sometimes returns a project item page with 0 items, a totalCount of 0 and hasNextPage: false, with no error. read_board accepted that page as the end of the board, so a run could see a board as shorter than it is, or as empty. On 22 Oct this happened often, on all 3 boards. This change makes the read detect such a page, try the read again, and stop the run when it stays short.

Changes

• read_board asks for items.totalCount and raises ShortBoardRead when a later page reports another total, when the number of items read differs from the total, or when the board has no items.
• A short board is read up to 6 times in all (5 retries), 10 seconds apart; when all 6 are short, the run stops with exit code 2 before any write. A board with no items therefore stops every run.
• README: the new rule, as its own item.

Validation

• python -m pytest -q: 171 passed.
• coverage run -m pytest -q && coverage report: 99% total, board_sync/github.py 100% of lines and branches.
• ruff check . and ruff format --check .: clean.
• I changed each new condition, the retry count, the wait, the retried error type and the totalCount field of the query in turn; every change makes at least 1 test fail.

Not checked

• The short pages were seen with a personal token; I did not reproduce them with the App token the workflow uses.
• Value lists inside an item (fieldValues, issueFieldValues) still rely on hasNextPage only.