Function clear_cursor_dirty 

pub async fn clear_cursor_dirty(
    pool: &SqlitePool,
    did: &str,
    feed_url: &str,
    flushed_updated_at: &str,
) -> Result<()>

Clear the dirty flag on a cursor after a successful PDS flush — but ONLY if the row still carries the exact flushed_updated_at snapshot we flushed.

The flusher reads a cursor, sends it to the PDS (a network round-trip), then clears dirty. A concurrent upsert_cursor (a fresh mark-read) can land DURING that in-flight write, bumping updated_at and re-setting dirty = 1 for reads that were NOT in the flushed snapshot. An unconditional SET dirty = 0 would silently drop those reads. Guarding on the snapshot’s updated_at makes this a compare-and-swap: if updated_at changed under us, zero rows update, the row stays dirty, and it re-flushes next round.