PagePool::{default -> new_for_test} + temporary hack for IN_MEMORY_CONFIG / test_index_scans

[Centril]

Description of Changes

First, PagePool::default is renamed to PagePool::new_for_test to make sure ::default() isn't used anywhere.
This was useful when debugging the issue.

Second, IN_MEMORY_CONFIG in crates/testing/src/modules.rs is changed so that page_pool_max_size: Some(size_of::<Page>()), which is the minimum possible. This makes the failures in test_index_scans go away. On my i7-7700K/64GB the threshold for the test failing seems to be at 1 << 26 but it passes on 1 << 25. As to why a page pool >= 64 MiB would cause such a slow down is unclear to me as of yet.

Third, the default page pool size is now Some(size_of::<Page>()) which effectively disables page pooling.

API and ABI breaking changes

None

Expected complexity level and risk

1

Testing

This fixes a test.

[gefjon]

• Needs companion PR to fix internal tests.
• Should we be using a smaller page pool everywhere, not just in tests? Do the page pool metrics have anything useful to tell us here?

The code changes look fine and I'm willing to approve this, but I'm not super excited to merge it without knowing whether using the same smaller pool everywhere would win us two orders of magnitude of performance somehow.

[Centril]

Needs companion PR to fix internal tests.

https://github.com/clockworklabs/SpacetimeDBPrivate/pull/1691

Should we be using a smaller page pool everywhere, not just in tests? Do the page pool metrics have anything useful to tell us here?

The metrics, if I'm reading them correctly, suggests that we were storing about 40k pages in the pool, whereas the limit is about 131k pages. Might be good to shrink to 4 GiB instead of 8 GiB as the default? But I would suggest that we do such changes in follow ups to unblock CI/other PRs.

[Centril]

We decided to disable the page pool in the host for now.

[Centril]

@coolreader18 the smoketests failing don't seem to have anything to do with this PR but look related to #2644?

[gefjon]

the smoketests failing don't seem to have anything to do with this PR but look related to #2644?

I think these are related to #2714 , actually.