HHH-20515 Fix findMultiple overwriting session state with cache values

[jjennnn]

Added check in CacheLoadHelper.convertCacheEntryToEntity() to return managed entities without reassembling from cache, preserving uncommitted changes. Added FindMultipleCacheTest that verifies findMultiple() keeps in-memory entity changes when called multiple times within the same session.

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license
and can be relicensed under the terms of the LGPL v2.1 license in the future at the maintainers' discretion.
For more information on licensing, please check here.

---

---

Please make sure that the following tasks are completed:
Tasks specific to HHH-20515 (Bug):

• Add test reproducing the bug
• Add entries as relevant to migration-guide.adoc OR check there are no breaking changes

https://hibernate.atlassian.net/browse/HHH-20515

[mbellade]

@jjennnn thanks for working on this, but I'm not sure this is the correct approach: changing the common 2nd level cache load path (which is used across many different methods, not only findMultiple()) to always return the persistent instance if one is found in the current session may be ok for ``findMultiple( ..., SessionCheckMode.ENABLED ), but it would conflict with the DISABLED` semantic. For the latter, we should always avoid looking in the PC and just return a (transient?) instance initialized with the values contained in the cache without refreshing its state.

Looking at the findMultiple() contract though, it says it returns persistent instances, and I'm not sure how we can avoid "refreshing" the state of an existing entity if we're supposed to skip checking the PC for it. @sebersole any insights on the intended behavior?

[mbellade]

@jjennnn your approach is still doing persistence context lookups upfront, even if the SessionCheckMode would be set to DISABLED. We need to avoid such lookups, and just return any existing persistent entity instance after the 2nd level cache lookup or the query occurred.

Edit: please add tests for both SessionCheckMode.DISABLED and ENABLED (passing it explicitly so we don't depend on the current default), as well as CacheMode.REFRESH_SESSION since your changes seem to account for that as well.

[jjennnn]

@mbellade I tried removing the oldHolder.getEntity() fallback and reran the tests, but CircularityTest#testSelection started failing. It looks like the reuse logic is still needed for circular graph initialisation, so I kept it.

[mbellade]

@jjennnn right, without the early return we probably need to maintain the lookup in CacheLoadHelper. The issue with that is, now we're doing a double map lookup in the negative case: when no holder is available, we are looking for one in both the multi-id entity loader and the 2nd level cache logic - with that said, the entityKey is pre-computed so that might be fine. The alternative would be to restore the early return in the generic 2nd level cache loader, and also return null there for deleted entries.

[sonarqubecloud]

Quality Gate failed

Failed conditions
79.6% Coverage on New Code (required ≥ 80%)

See analysis details on SonarQube Cloud