Performance: Redundant catalog read in `pending_sync_rows()` which increases fetch time

[aditya3799]

Background

In a previous optimization (PR #19), a redundant call to selected_or_default_graph_id_via_definer() was successfully removed from current_catalog_state(). This helped reduce the ~3.5ms unamortized fixed cost of graph traversals.

The Problem

However, looking closely at ensure_current_graph(), the very next line calls pending_sync_rows(applied_sync_id).

Tracing this call shows that it eventually hits SyncReplayContext::load(), which triggers a full read_catalog(). This means ensure_current_graph() is still doing a full redundant catalog read (including two massive SPI queries for _registered_tables and _registered_edges), completely undermining the optimization from PR #19.

Proposed Solution

Since ensure_current_graph() already loads catalog_state right before calling pending_sync_rows(), we can extract the applicable_table_oids from the loaded state and pass them directly to pending_sync_rows. This bypasses SyncReplayContext::load() entirely and eliminates the remaining redundant SPI queries from the hot path.

I will submit a PR with this fix shortly!