chore: suspension type validation refactor

[nayib-jose-gloria]

Reason for Change

• Suspension type implementation introduced needless complication into schema definition yaml and was not performant, in exchange for more detailed error messages
• I don't think this trade-off has been worthwhile, so I propose simplifying the suspension type validation to follow patterns other fields' validation use and improve performance by reducing the total number of pandas df queries the validator runs.
• TODO: Refactor loses ability to warn but still accept unrecognized assays. Must account for that

Changes

• remove "complex_rule" pattern in schema_definition.yaml, use 'dependencies -> rule' pattern
• remove "complex_rule" implementation in valdiate.py (designed to be extensible, but no new field using it has come up in the last two years)
• group queries by suspension type (3) rather than by assay type (25 and counting)
• use slightly more generic error messages to accommodate this pattern (instead of specifying what suspension_types are valid for a failing assay x suspension_type combo, just flags that it is an invalid combo and to check the schema for a valid suspension type for the given assay type)
• small refactor to schema_definition.yaml to list all dependent validation rules for a field under "dependencies" rather than supporting both a list under 'dependencies' and a "default" validation defined outside of "dependencies". This affected self_reported_ethnicity_ontology_term_id and development_stage_ontology_term_id.
• small refactor to "_validate_column" command to account for removing "default" dependency behavior.

Testing

• unit tests for regression testing
• TBD notebook testing for performance improvement

Notes for Reviewer