Fix flaky cagg_insert

Isolation test `cagg_insert` is flaky because refresh steps can
complete in any order. Adding constraints so that completion is
reported in the same order in all test runs.

tsl/test/isolation/specs/cagg_insert.spec

@@ -123,15 +123,15 @@ step "LockMatInval" { BEGIN; LOCK TABLE _timescaledb_catalog.continuous_aggs_mat
 step "UnlockMatInval" { ROLLBACK; }
 
 #only one refresh
-permutation "LockInvalThrEx" "Refresh" "Refresh2" "Refresh3" "UnlockInvalThrEx"
+permutation "LockInvalThrEx" Refresh Refresh2(Refresh) Refresh3(Refresh,Refresh2) "UnlockInvalThrEx"
 
 #refresh and insert do not block each other once refresh is out of the
 #first transaction where it moves the invalidation threshold
 permutation "Ib" "LockCagg1" "I1" "Refresh" "Ic" "UnLockCagg1"
 permutation "Ib" "LockCagg1" "Refresh" "I1" "Ic" "UnLockCagg1"
 
-permutation "I2b" "LockCagg2" "I21" "Refresh" "Refresh3" "I2c" "UnLockCagg2"
-permutation "I2b" "LockCagg2" "Refresh3" "Refresh" "I21" "I2c" "UnLockCagg2"
+permutation "I2b" "LockCagg2" "I21" Refresh Refresh3(Refresh) "I2c" "UnLockCagg2"
+permutation "I2b" "LockCagg2" Refresh3 Refresh(Refresh3) "I21" "I2c" "UnLockCagg2"
 
 #refresh and select can run concurrently. Refresh blocked only by lock on
 # cagg's materialized hypertable. Needs RowExclusive for 2nd txn.
@@ -159,5 +159,5 @@ permutation "Refresh" "SV1" "LockMatInval" "Refresh1" "Ib" "I1" "LockInvalThrEx"
 permutation "I1" "Refresh" "LockInval" "Refresh" "Sb" "S1" "Sc" "UnlockInval"
 permutation "I1" "Refresh" "LockInval" "Sb" "S1" "Refresh" "Sc" "UnlockInval"
 
-permutation "I1" "I21" "Refresh1" "Refresh2" "Refresh3"
-permutation "I1" "I2b" "I21" "Refresh2" "Refresh3" "I2c" "Refresh3"
+permutation "I1" "I21" Refresh1 Refresh2(Refresh1) Refresh3(Refresh1,Refresh2)
+permutation "I1" "I2b" "I21" Refresh2 Refresh3(Refresh2) "I2c" Refresh3(Refresh2)