Improve performance with efficient use of FieldMappingCache

[V1NAY8]

This small change will improve the performance of FieldMappingCache

Existing drawbacks:

• field_mapping_cache is initialized on every yield of itertuples and iterrows
• Whenever eland dataframe is used to repr() or print() the field_mapping_cache is re initialized on every call.

Now, field_mapping_cache is initialized once per query_compiler, that too only when ETL such as to_pandas() or itertuples or iterrows or repr() or print() is done.

Also, by default scripted_fields are considered as "object". Previously we were finding source field, then if it's not source field, an exception is raised and then its data type is assigned as "object".

This also improves the readability of code.

@sethmlarson Take a look :)

[elasticmachine]

Since this is a community submitted pull request, a Jenkins build has not been kicked off automatically. Can an Elastic organization member please verify the contents of this patch and then kick off a build manually?

[sethmlarson]

jenkins test this please

[sethmlarson]

jenkins test this please

[V1NAY8]

Ok, So, the tests are good. So, a simple rebase should be fine.

[sethmlarson]

jenkins test this please