- add check for the two cases in table stats service:
  - "large enough" row buffer (currently "large enough" is anything more than maxDeltaRows/8)
    that has not seen any updates/deletes since the last check; in this case schedule
    a task to force rollover the row buffer in column table
  - also check if a bucket of column table has multiple small batches (non-transactional check);
    if so then submit a task to merge those after checking for transactional snapshot;
    merge is done by locally created ColumnTableScan->ColumnInsertExec plan where the scan
    uses an iterator only on the small batches
- added a ColumnFormatStatsIterator that can take a bunch of stats rows and create an iterator
  over just those (like required for batch merge)
- added new scan metrics for disk reads: a) disk rows from row buffer,
  b) partial column batches on disk, c) full column batches on disk
- extended SQLMetrics types with a new SPLIT_SUM_METRIC that allows displaying multiple metrics
  against a common name; ColumnTableScan now uses this to combine some metrics else it becomes
  too large in display (especially for the newly added disk read metrics)
- use hive-metadata (ExternalTableMetaData) to get number of rows instead of getting
  from row buffer table (that is subject to changes in future)

- fixed disk metrics collection added previously; set the metric correctly
  for both row buffer iterator (ResultSetTraversal) and ColumnFormatIterator
- added a metric for remote batch fetch
- fixed multiple ColumnTableScans causing split metrics to add up into one
  ColumnTableScan; now use a unique ID for split metrics for each ColumnTableScan instance
- fix an NPE in SnappyTableStatsProviderService while filling up
  result map from members since CHM cannot hold null values
- use a common entry map in ColumnFormatIterator disk iteration instead of creating
  separate for every column batch
- added implementation of PURGE_CODEGEN_CACHES as StoreCallbacksImpl.clearCodegenCaches
- limit to one task per table for background rolloverRowBuffer and mergeSmallBatches tasks
- replaced a few usage of Map.put with justPut for koloboke maps

… replicated table

Conflicts:
	core/src/main/scala/io/snappydata/SnappyTableStatsProviderService.scala
	core/src/main/scala/io/snappydata/TableStatsProviderService.scala
	store

Conflicts:
	cluster/src/test/scala/io/snappydata/benchmark/TPCHColumnPartitionedTable.scala

Conflicts:
	core/src/main/scala/org/apache/spark/sql/store/CompressionCodecId.scala

…el with update/delete

added an OperationContext in SnappySession that can be used to persist context across
multiple plan executions (e.g. caching for putInto, then actual execution)