[WIP] switch to parallel queries, and write intermediate dataframes to disk

[akotlar]

• Switch get_annotation_result_from_query to parallel executor
• Save memory by writing intermediate dataframes to disk, and reading in the data as memory mapped file

WIP. Tests still need to be written, and schema must be provided explicitly.

# Default behavior is to show 1 sample per row, e.g. `melt_samples=True`
query_result_df_array_of_structs_select_fields = get_annotation_result_from_query(
    query_string="cadd:>20",
    index_name=index_name,
    bystro_api_auth=user,
    fields=['refSeq.name2', 'refSeq.name', 'refSeq.exonicAlleleFunction', 'refSeq.siteType'],
    explode_field='refSeq.name2',
    tmp_dir='/mnt/ssd1/bystro/python/python/bystro/examples/tmp',
    output_path='foo'
)
query_result_df_array_of_structs_select_fields.head(n=10)

Here tmp_dir and output_path are optional. If these are not provided the intermediate files as well as the concatenated final table will be cleaned up after completion.