cleanup connection handling

[luke-lombardi]

Summary by cubic

Simplified pod connection tracking by removing Redis aggregates and relying only on per-proxy snapshot counters with TTL. This avoids stale counts and ensures accurate shared totals across proxies for autoscaling.

• Refactors
  • Removed aggregate keys and fallbacks (Keys.podTotalConnections, Keys.podContainerConnections); totals are now computed by summing snapshot keys (Keys.podProxyConnections*).
  • flushConnectionState publishes only snapshot counters with TTL; aggregate refresh path deleted.
  • Updated autoscaler and proxy tests to read shared counts from snapshots and verify stale aggregates are ignored; removed the unused podContainerConnectionTimeout.

^{Written for commit 6298c7d. Summary will update on new commits.}