fix(profiler): query system tables via clusterAllReplicas

Summary

• Query system.query_log, system.processors_profile_log, and other system tables via clusterAllReplicas('default', ...) so lookups find entries regardless of which node executed the profiled query.
• Handle string-encoded numeric values in JSON responses from clusterAllReplicas (Nullable column wrapping causes ClickHouse to serialize integers as strings in JSONEachRow format).

Without this fix, --profile silently returns zero for all stats (read_rows, memory_usage, etc.) when the system table query lands on a different node than the one that ran the profiled query. !786 (merged) requires this for profiling.