tests: make reconcile_until_idle more aggressive by default

[jcsp]

Problem

The reconcile_until_idle helper backs off when the controller appears to be busy. However, this leads to long runtimes when there are multiple optimisations pending, for example after a shard split. The #9916 changes made migrations take extra steps in some cases (creating temporary secondary locations), which highlights runtime impact of reconcile_until_idle.

Example failure of test_timeline_ancestor_detach_idempotent_success where reconcile had backed off to 5s period
https://neon-github-public-dev.s3.amazonaws.com/reports/pr-10411/12794755547/index.html#/testresult/47e461aab29cdf4b

Summary of changes

• Change the default max backoff to 1s from 5s. This means that in the default 30s timeout we should have comfortably enough iterations to quiesce an 8 shard split, so that tests doing this don't have to pass any non-default arguments. Tests that do huge numbers of operations will continue to override max_interval as needed.

[github-actions]

6660 tests run: 6305 passed, 0 failed, 355 skipped (full report)

---

Flaky tests (4)

Postgres 17

• test_pageserver_metrics_removed_after_detach: release-arm64

Postgres 16

• test_pull_timeline_while_evicted: release-x86-64

Postgres 15

• test_readonly_node_gc: release-x86-64

Postgres 14

• test_physical_replication_config_mismatch_max_locks_per_transaction: release-arm64

Code coverage* (full report)

• functions: 33.7% (8422 of 25017 functions)
• lines: 49.2% (70466 of 143315 lines)

* collected from Rust tests only

---

_{The comment gets automatically updated with the latest test results
9056c7d at 2025-01-16T09:56:44.206Z :recycle:}