#517 [API/Memory] DataShard table unbounded memory loading

**File:** api/blueprints/v1/logs.py:90-119 **Problem:** `_read_table_to_pandas()` loads ENTIRE DataShard tables into memory on every request. With continuous workflow execution, these tables grow indefinitely. No pagination at DataShard level. **Severity:** CRITICAL **Fix:** - Implement LRU cache with TTL for table snapshots - Add predicate pushdown to DataShard queries - Implement streaming/chunked reading - Add memory limits per-query **Impact:** Production systems with high workflow volume will exhaust memory.