Use INSERT OR IGNORE for excluded_statuses markers

[mjhfunctionalashtanga]

Problem

During refreshUnreadEntries → markAllUnread, syncing fails with:

android.database.sqlite.SQLiteConstraintException: UNIQUE constraint failed:
excluded_statuses.article_id, excluded_statuses.type (code 1555)
    at com.jocmp.capy.db.ArticlesQueries.upsertUnread

upsertUnread / upsertStarred insert a marker row into excluded_statuses(article_id, type) with no conflict clause (the sibling article_statuses insert correctly uses ON CONFLICT). markAllUnread runs all the marker inserts and the updateStaleUnreads() cleanup — which deletes those markers — in a single transaction. So if any marker already exists (a duplicate ID in the batch, or a marker stranded by an earlier interrupted sync), the insert throws, the whole transaction rolls back, the cleanup never runs, and the leftover markers guarantee the same failure on every subsequent sync. Unread refresh never completes and the client churns.

Fix

Make the marker insert idempotent with INSERT OR IGNORE. Duplicates/leftovers are skipped, the transaction completes, and updateStaleUnreads / updateStaleStars clean up the markers as intended — self-healing.

Verified on a Boox device (Android 11): the recurring db_error disappears and unread refresh completes cleanly.

🤖 Generated with Claude Code

https://claude.ai/code/session_015yVKgdxd3ZSrFeY5eWpFuA

[jocmp]

Pushed a follow-up commit to this branch (maintainer edit) rather than opening a competing PR.

Two additions on top of the INSERT OR IGNORE:

• .distinct() at the markAllUnread/markAllStarred boundary. The duplicate originates upstream — fetchAllEntryIDs (Miniflux) and fetchAllItemIDs (Reader) paginate by offset against total with no dedup, so a set shifting mid-pagination can yield the same ID on two pages. Worth stopping there too, not only at the insert.
• Three regression tests: duplicate IDs through both markAll* entry points, plus one driving upsertUnread twice inside a transaction so the SQL change stays covered independently of the .distinct().

Verified the tests fail with either fix reverted and pass with both — :capy:test --tests ArticleRecordsTest, 27 tests.

One note on the description: the "marker stranded by an earlier interrupted sync" mechanism doesn't hold, since the marker inserts and updateStaleUnreads() share a transaction and a rollback takes the markers with it. The duplicate-ID-in-batch path is real on its own and is what the fix addresses.