fix(mobile): retain live thread replies in channel window

[smltr]

Summary

Mobile received ordinary live thread replies but discarded them at the channel-window storage boundary unless they carried broadcast=1. Because the replies never reached the window store, the parent message could not update its thread summary until the app reloaded the channel from the relay.

This change:

• retains live non-broadcast thread replies in the mobile channel-window store
• leaves the existing render-layer filter responsible for keeping ordinary replies out of the top-level timeline
• removes the now-unused store-layer broadcast helper
• adds regression coverage proving a live reply increments the parent reply count without becoming a top-level row

Related issue

Fixes #3293.

An issue/PR search found no duplicates

Testing

• flutter test test/features/channels/channel_messages_provider_test.dart — 15 passed
• just mobile-check — formatting, Flutter analysis, and mobile file-size checks passed
• flutter test — 826 passed, 1 skipped
• git diff --check — passed
• before/after smoke test on a my Pixel 8a:
  • Baseline 2ce2d71cc: an incoming threaded reply did not update the parent summary until the app was restarted.
  • Fix 0f7fec6ca: the parent gained the reply summary live without restart, and the reply remained out of the top-level timeline.

No screenshots or recordings captured

[idrisskone101]

bumping this 👆🏾

[daguiler76]

Current main introduced a content conflict in channel_messages_provider_test.dart after this PR's last update. I prepared a minimal follow-up against the PR head that only reorders existing tests so Git retains both main's new local-row regression and this PR's two live-reply regressions:

smltr#1

Verified on the follow-up head:

• merge simulation with current main: clean (git merge-tree --write-tree, exit 0)
• just mobile-check: passed
• full mobile suite: 826 passed, 1 skipped

I also resolved the integration directly on current main as a separate cross-check; that combined tree passed 1,024 tests with 1 skipped. No production behavior changed in the helper commit.