Implement pydantic validation of existing settings file

[bendnorman]

Requirements:

• [x] The new ETL settings file should allow us to specify inputs for both the FERC1 DB cloning and the main PUDL ETL, with the possibility of adding new sections as we incorporate additional datasets and/or processes.
• [x] When enumerating the datasets to be processed, use a dictionary with codes (epacems, eia, ferc1 etc.) not a list.
• [x] The validator should be aware of what codes represent valid datasets, and ensure that only those codes are found.
• [x] Standardize on using eia860m rather than eia860_ytd as a code for that eia sub-dataset.
• [x] The years, states, and any other partitions of the data should be sorted after input so it is easy to track progress through the ETL.
• [x] We may want to include default values so that not everything needs to be explicitly spelled out in the intput file. Or we can use a special keyword like all which causes all working data partitions to be processed.
• [x] We may also infer other implicit tables or datasets that need to be added, in order to make everything else work (the glue tables and static dataframes).
• [x] Remove or rename any remaining references to datapackages in the settings
• [x] Make sure the validator provides helpful error messages when something goes wrong.
• [x] yml files in package_data/settings
• [x] #1324
• [ ] Remove the PUDL_TABLES and WORKING_PARTITIONS dictionaries from pudl.constants as that information will be subsumed within the ETL settings class / validation structures.
• [ ] Ensure that only working partitions of the data are specified. Currently this is defined by the tables, years, and states found in PUDL_TABLES and WORKING_PARTITIONS. Note that this is potentially distinct from what raw input data is available, and that information is already stored in the metadata of the raw input datapackages.

[bendnorman]

1292 closed out 10/13 of these tasks.

I created #1324 for the unique dataset problem. Pretty bizarre that PyYaml doesn't enforce unique keys.

The two remaining tasks are related to where we specify working tables and partitions. I decided to keep PUDL_TABLES and WORKING_PARTITIONS dictionaries in pudl.constants because it was cleaner than including these constants in the pydantic settings models. I suppose we could infer working partitions from datapackage.jsons. This could be an issue because we might archive all partitions of a dataset but only implement the ETL for some of the partitions. If we infer the partitions from datapackage.json we would need to specify missing partitions in the code somewhere. This is discussed in #1264.