[SPARK-12937][SQL] bloom filter serialization

[cloud-fan]

This PR adds serialization support for BloomFilter.

A version number is added to version the serialized binary format.

[cloud-fan]

bloom filter and count-min sketch can have different version values, but we can share same version class.

[rxin]

I think we should move it back, because:

1. The version enum is actually the best place to document the binary protocol.
2. This will be really confusing when bloomfilter has v2 and yet count-min sketch has only v1.
3. The amount of code duplication you save is teeny (actually you probably added more loc by having an apache licensing header).

[rxin]

cc @liancheng on point 1 - the best place to document the binary protocol is in Version!

[cloud-fan]

cc @rxin @liancheng

[rxin]

also include the current value

[SparkQA]

Test build #50081 has finished for PR 10920 at commit 4b05a35.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[cloud-fan]

cc @liancheng , I removed this as the design doc says users should not care about the version being used.

[SparkQA]

Test build #50084 has finished for PR 10920 at commit 38d674c.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #50086 has finished for PR 10920 at commit c9b29c9.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[rxin]

I'm going to merge this first. Please move the num hash function thing in your next pr. Thanks.