[Sync] Refactoring polling to be reliable.

Polling is now an important component of sync health, as it can sometimes be
the only time we'll query for certain datatypes. As such, the following
has changed:
- Polls that fail will be retried (with backoff).
- As such, the logic to force a poll after an auth error isn't needed anymore
- The last successful poll time will be persisted in the sync prefs.
- On startup, schedule the first poll for last_poll_time + poll_interval
  (or Now(), whichever is latest).
- Receiving a new poll interval from the server will update the poll timer
- The poll timer is now a one shot timer, and only restarts on success
- Some code cleanup to make the above more straightforward

BUG=482154

Review URL: https://codereview.chromium.org/1132013004

Cr-Commit-Position: refs/heads/master@{#329669}

1 files changed, 1 insertions, 1 deletions

diff --git a/sync/internal_api/test/fake_sync_manager.cc b/sync/internal_api/test/fake_sync_manager.cc
index 43b6eb6..94ebdac 100644
--- a/sync/internal_api/test/fake_sync_manager.cc
+++ b/sync/internal_api/test/fake_sync_manager.cc
@@ -118,7 +118,7 @@ void FakeSyncManager::UpdateCredentials(const SyncCredentials& credentials) {
 }
 
 void FakeSyncManager::StartSyncingNormally(
-      const ModelSafeRoutingInfo& routing_info) {
+      const ModelSafeRoutingInfo& routing_info, base::Time last_poll_time) {
   // Do nothing.
 }