[Lake] Integrate pdr_subscriptions into GQL Data Factory

[idiom-bytes]

Motivation

Subscriptions are not being fetched from subgraph and added to lake. There is similar logic being used to determine consumptions for df_buyer, but the requirements are different.

Instead, simply use predictSubscriptions w/ a top-level timestamp filter to fetch all subscriptions between the expected [st_ut, fin_ut].

When implementing this, we'll have 2 different subgraph queries being handled by GQL Data Factory, using the same date-range. Due to subgraphs tables having records (or not), the Data Factory should elegantly handle empty gql quries, empty parquet files, and it's basic interface of get_dfs() without any issues.

DoD:

• Improve gql_data_factory so it can handle empty fetches, empty parquet files, and general mis-matches between fs.
• Improve gql_data_factory to return an empty dataframe of the expected schema if nothing is availbale.
• Add fn to fetch subscriptions from subgraph
• Create subscription object for models
• Create subscription schema for lake
• Hook subscriptions to lake
• Add mock utils for subscriptions
• Add tests that verify that subscription objects, subgraph, and components are working as expected
• Add tests that verify that GQL Data Factory that handle empty subgraph fetches + empty parquets files

[idiom-bytes]

Implemented pdr_subscriptions while updating GQL Data Factory, such that it behaves in the expected manner.

Example:

1. gql_data_factory already has predictions.parquet saved locally and up-to-date, so it skips
2. gql_data_factory doesn't have subscriptions.parquet, so it tries to update it
3. gql_data_factory can't find subscriptions, so it doesn't save to parquet
4. gql_data_factory loads predictions.parquet successfully, creating charts
5. gql_data_factory can't load subscriptions.parquet, so it returns an empty dataframe w/ the expected schema