[SNAP-3165] Instantiating snappy session only when catalogImplementation

[vatsalmevada]

is in-memory which running pyspark shell.

What changes were proposed in this pull request?

We are initializing SparkSession as well as SnappySession while starting pyspark shell.
SparkSession and SparkContextwere always initialized with hive support enable
irrespective of value of spark.sql.catalogImplementation config.

With these changes, we are checking the value of spark.sql.catalogImplementation and
hive support is not enabled when the value of above-mentioned property is set to
in-memory explicitly.

SnappySession will be only initialized when catalog implementation is set to in-memory
to avoid failure reported in SNAP-3165.

Later we can provide support for hive catalog implementation for python with SnappySession.

How was this patch tested?

manual

Following scenarios are tested:

• Created external table from pyspark shell and checked that it is visible
  on the embedded cluster UI.

>>> snappy.sql(" create external table airline using parquet options(path '/path/to/airline_withId_210M');")

• Restarted pyspark shell and was able to access the earlier created external
  table

>>> snappy.sql("select * from airline limit 10").show()

• Created column table from pyspark and checked that it is visible
  on the embedded cluster UI.

>>> snappy.sql("create table snappy_table(id int, name string) using column") 
>>> snappy.sql("insert into snappy_table values(1,'abc')")

• Restarted pyspark and was able to access column table

>>> snappy.sql("select * from snappy_table").show()

• Started spark-shell and was able to access external table as well as column
  table

[suranjan]

LGTM.
We need to document this behaviour.