validate_shards is too slow on large datasets — eats seconds before every SQL query

[tig888]

wsds.utils.validate_shards is called eagerly inside WSDataset.sql_select, to "warm up the filesystem cache" before any read. On a 7010-shard dataset on Weka with ~6 column dirs, this takes ~21s per call, dominating query latency for any workload that opens the dataset.