test: Give testConcurrentIncrementsWithSelectForUpdate room to retry on MariaDB

[Oleg Babichev (obabichevjb)]

Description

Summary of the change: EntityCacheRefreshTests.testConcurrentIncrementsWithSelectForUpdate
fails intermittently on CI against MariaDB with error 1020 (ER_CHECKREAD); this gives its
transactions enough retry attempts and a randomised backoff to absorb the conflict that MariaDB's
innodb_snapshot_isolation now raises by design.

Detailed description:

• Why: MariaDB has defaulted innodb_snapshot_isolation to ON since 11.6.2 (MDEV-35124), and
  the CI container reports 12.1.2-MariaDB with the setting ON. Under it, a locking read of a row
  that has changed since the transaction's snapshot no longer blocks and re-reads — it raises
  ER_CHECKREAD, "Record has changed since last read", and rolls the transaction back. MariaDB's
  guidance is to restart the transaction.

  This test is maximally exposed to that, and by its own design. Its first statement is a deliberate
  plain read whose only purpose is to make the DAO cache a stale value, and that read is exactly
  what establishes the transaction's snapshot:

  // This is important line, because it forces DAO to cache
  // the value from the beginning of transaction
  TestEntity.find { TestTable.id eq entityIdValue }.single()

  The forUpdate() read immediately after is the locking read. With 20 transactions incrementing a
  single row, another transaction committing inside that window is routine rather than exceptional.
  So the mechanism the test uses to prove the DAO refreshes its cache is the one that snapshot
  isolation rejects.

---

Type of Change

Please mark the relevant options with an "X":

• Bug fix

Affected databases:

• MariaDB