[SPARK-11191] [SQL] [FOLLOW-UP] Cleans up unnecessary anonymous HiveFunctionRegistry

[liancheng]

According to discussion in PR #9664, the anonymous HiveFunctionRegistry in HiveContext can be removed now.

[SparkQA]

Test build #45997 has finished for PR 9737 at commit 1cb40df.

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

[yhuai]

Should we just pass in executionHive? We do not use other fields of HiveContext, right?

[liancheng]

@yhuai I think metadataHive is also required if we want to handle persisted functions correctly later.

[yhuai]

Then, should we add that later? I am thinking passing in HiveContext may be not really necessary if we only to use one or two fields.

[liancheng]

Discussed with @yhuai offline. His major concerns are:

1. Currently there's are circular references between HiveContext and HiveFunctionRegistry, which isn't obviously harmful for now, but would be good to eliminate.
2. It's not necessary to pass the whole HiveContext into HiveFunctionRegistry when we only access 1 or 2 fields.

I agree. Updating.

[yhuai]

LGTM pending jenkins.

[SparkQA]

Test build #46058 has finished for PR 9737 at commit cd2b958.

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

[liancheng]

Merged to master and cherry-picked to branch-1.5.

[liancheng]

Also backported to branch-1.6.