test flake: refresh_begins_at_the_committed_frontier_after_background_summaries_publish times out under load #4518

Observed

//crates/tracedecay-session-runtime:unit_test intermittently fails on PRs that do not touch session or LCM code:

test lcm_effects::tests::refresh_begins_at_the_committed_frontier_after_background_summaries_publish ... FAILED
panicked at crates/tracedecay-session-runtime/src/lcm_effects/tests.rs:1647:5

Seen on PR #4478 (Renovate MCP client bump) and PR #4510 (code-index cold-pass retry, run 40087310818). The test waits about 30 s in an eventually loop for background summaries to publish before a refresh begins. Under loaded CI runners the background summary publication does not finish in time.

Expected

The test asserts the ordering it cares about (the refresh starts at the committed frontier after summaries publish) through a deterministic signal from the summary publisher, not through a wall-clock wait, so runner load cannot fail it. If the wait is hiding a real ordering bug, find and fix it instead.