Fix FeedListViewModelTest flakiness at its actual root cause (issue #261)

[mapitman]

Closes #261

Summary

Root-caused rather than contained (unlike the sibling ArticleListViewModelTest/SettingsViewModelTest flakiness, issues #54/#60/#215, where three separate root-cause attempts failed and a CI-only class skip was the only thing that stuck -- see repo history). Downloaded the actual CI failure artifacts from the run that failed this issue and found two distinct, understandable bugs:

• uiState_splitsPodcastAndOtherFeedsIntoFixedSections failed with expected:<[2]> but was:<[]> -- its wait predicate was sections.any { isNotEmpty }, which only requires one of the two sections to have settled. Room delivers each subscribed feed's Flow update from its own background invalidation-tracker thread, so the sections can populate at different times, and the test could assert against a still-settling state. Fixed to wait for both feeds to have landed (sections.sumOf { it.feeds.size } == 2).
• uiState_freezesUnreadCountWhileRefreshing failed with expected:<1> but was:<2> -- it (and a sibling test, uiState_allReadFeedWithNoNewItems_unreadCountNeverRisesDuringRefresh, same pattern) used a real 300ms MockResponse delay to keep the refresh coroutine genuinely in-flight while the test writes to the DB mid-refresh, racing real wall-clock time against the virtual test dispatcher -- exactly the failure mode a loaded CI runner exposes. Replaced with a MockWebServer Dispatcher blocked on a CountDownLatch the test releases explicitly instead of a real network response body reaching completion after a hoped-to-be-long-enough delay: same real dispatcher-switch mechanics (the network call still genuinely suspends on Dispatchers.IO), no more racing a fixed duration.

Test plan

• FeedListViewModelTest in isolation, 10/10 clean reruns
• Full testDebugUnitTest suite, 3/3 runs -- FeedListViewModelTest 100% clean each time (the one unrelated failure each run was the already-tracked ArticleListViewModelTest class, issues Investigate CI-only hang in SettingsViewModelTest #54/Investigate CI-only failure in ArticleListViewModelTest.defaultToAllArticleViewSetting_showsAllByDefault #60/ArticleListViewModelTest: refresh() corrupts next test's StateFlow wait (lost wakeup) #215)
• Actual CI run (per project policy, local pass streaks alone aren't proof of a CI-only flake being fixed)