Describe the problem

The workload expects an error whenever a schema change is performed after a write. This logic is incorrect because the following succeeds:

root@localhost:26257/defaultdb  OPEN> create table foo (i int, j int);
CREATE TABLE

root@localhost:26257/defaultdb  OPEN> create table bar (j int);
CREATE TABLE

root@localhost:26257/defaultdb> begin;
BEGIN

root@localhost:26257/defaultdb  OPEN> alter table bar drop column j;
ALTER TABLE

root@localhost:26257/defaultdb  OPEN> insert into foo values (1,2);
INSERT 1

# This DML succeeds even though the previous statement was a write. 
root@localhost:26257/defaultdb  OPEN> create table t();
CREATE TABLE

root@localhost:26257/defaultdb  OPEN> commit;
COMMIT

Expected behavior

As long the first DDL precedes the first write in a txn, then you are allowed to perform schema changes after writes in the txn. The workload should be updated to follow this rule instead of simply expecting an error for every DDL that occurs after a DML.

Additional data / screenshots

This error has shown up in these roachtest/schemachange/random-load failures:
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56081 (comment)
#56235 (comment)
#56230 (comment)
#56081 (comment)
#56081 (comment)
#56235 (comment)
#56230 (comment)
#56081 (comment)
#56081 (comment)
#56230 (comment)

Jira issue: CRDB-3428