fix: prevent read entries from reverting to unread on re-fetch

Root cause: cleanup_stale_entries deleted read-but-unsaved entries from
active feeds, then the Go worker re-inserted them with new UUIDs,
orphaning the user_entry_states rows and making entries appear unread.

- cleanup_stale_entries: skip feeds with active subscribers and preserve
  entries that have been read (not just saved)
- Go parser: normalize GUIDs by trimming whitespace and stripping
  tracking query parameters from URL-based identifiers
- Go writer: preserve original published_at on upsert instead of
  overwriting, preventing old entries from jumping to timeline top
- get_unread_counts: apply same time boundary as get_timeline so
  ancient re-inserted entries don't inflate counts
- Realtime listener: ignore INSERT events for entries older than 48h
  to suppress misleading "new entries" notifications from re-inserts

1 files changed, 2 insertions, 0 deletions

diff --git a/services/worker/internal/fetcher/errors_test.go b/services/worker/internal/fetcher/errors_test.go
index e81251b..5c425aa 100644
--- a/services/worker/internal/fetcher/errors_test.go
+++ b/services/worker/internal/fetcher/errors_test.go
@@ -107,7 +107,9 @@ func TestClassifyTimeoutError(test *testing.T) {
 type timeoutErr struct{}
 
 func (timeoutError *timeoutErr) Error() string   { return "connection timed out" }
+
 func (timeoutError *timeoutErr) Timeout() bool   { return true }
+
 func (timeoutError *timeoutErr) Temporary() bool { return true }
 
 func TestClassifyNetworkOpError(test *testing.T) {