[SPARK-1402] Added 3 more compression schemes

JIRA issue: [SPARK-1402](https://issues.apache.org/jira/browse/SPARK-1402)

This PR provides 3 more compression schemes for Spark SQL in-memory columnar storage:

* `BooleanBitSet`
* `IntDelta`
* `LongDelta`

Now there are 6 compression schemes in total, including the no-op `PassThrough` scheme.

Also fixed a bug in PR #286: not all compression schemes are added as available schemes when accessing an in-memory column, and when a column is compressed with an unrecognised scheme, `ColumnAccessor` throws exception.

Author: Cheng Lian <lian.cs.zju@gmail.com>

Closes #330 from liancheng/moreCompressionSchemes and squashes the following commits:

1d037b8 [Cheng Lian] Fixed SPARK-1436: in-memory column byte buffer must be able to be accessed multiple times
d7c0e8f [Cheng Lian] Added test suite for IntegralDelta (IntDelta & LongDelta)
3c1ad7a [Cheng Lian] Added test suite for BooleanBitSet, refactored other test suites
44fe4b2 [Cheng Lian] Refactored CompressionScheme, added 3 more compression schemes.