Master rebase, fix timestamp inference.

Bugfixes

• Fix broken timestamp rebase to do automatic schema inference on timestamp types correctly (#76)

Improvements

• Merge latest from upstream master (#73)

InSet fix and timestamp pushdown

Breaking Changes

• The Timestamp type is now written into Parquet as an INT64 column instead of the previous INT96 type. This version of Spark still supports reading data from Parquet files written in the previous INT96-serialized format, but older versions of Spark will not be able to read Parquet files with the Timestamp type written by this Spark version.

Improvements

• Timestamp and date filter pushdown (#72)
  • Enabled due to the INT96 -> INT64 serialization change above.

Bugfixes

• Fix inset for large queries. Disable record level filtering (#74)

Fix broken parquet splitting

Improvements

• Resync with upstream master (#64)

Bug fixes

• Fix createFilter for NOT queries (#68)
• Fix for ineligible filters, use compressed block size (#70)

2.1.0-palantir18

DO NOT USE: this release contained a correctness regression that is fixed in 2.1.0-palantir19

• SPARK-17059: Allow FileFormat to specify partition pruning strategy (#62)
• open reader only once with correct schema (#59)
• make dist land in maven style directory pattern (#57)

2.1.0-palantir17

• Republish same code as 2.1.0-palantir16

2.1.0-palantir16

• Rework the Parquet partition pruning feature (#52)

2.1.0-palantir15

• hotfix an issue with the merge from apache master done in 2.1.0-palantir14

2.1.0-palantir14

• toLocalIterator on CollectLimitExec doesn't shuffle (#46)
• toJson doesn't perform toRdd conversion (#48)
• parquet bump to 1.9.0-palantir3 (#49)
• resync with apache master

2.1.0-palantir13

• Bump parquet version to 1.9.0-palantir2 (#45)
• Resync with apache master

2.1.0-palantir12

Merge branch 'master' into palantir-master