storcon: reproduce shard split issue (#11290)

## Problem

Issue https://github.com/neondatabase/neon/issues/11254 describes a case
where restart during a shard split can result in a bad end state in the
database.

## Summary of changes

- Add a reproducer for the issue
- Tighten an existing safety check around updated row counts in
complete_shard_split

storage_controller/src/persistence.rs

@@ -997,10 +997,11 @@ impl Persistence {
                 // Clear sharding flag
                 let updated = diesel::update(tenant_shards)
                     .filter(tenant_id.eq(split_tenant_id.to_string()))
+                    .filter(shard_count.eq(new_shard_count.literal() as i32))
                     .set((splitting.eq(0),))
                     .execute(conn)
                     .await?;
-                debug_assert!(updated > 0);
+                assert!(updated == new_shard_count.count() as usize);
 
                 Ok(())
             })

storage_controller/src/service.rs

@@ -5456,6 +5456,8 @@ impl Service {
             }
         }
 
+        pausable_failpoint!("shard-split-pre-complete");
+
         // TODO: if the pageserver restarted concurrently with our split API call,
         // the actual generation of the child shard might differ from the generation
         // we expect it to have.  In order for our in-database generation to end up