[ON HOLD] feat: auto detection of model encoding types

[shuangwu5]

Main changes:

• Introduce new model encoding types: TABULAR_AUTO, LANGUAGE_AUTO
• Change the behavior of the existing model encoding type AUTO: Before the PR, it means the same as TABULAR_AUTO, now it is like the union of TABULAR_AUTO and LANGUAGE_AUTO
• Extend auto detection logics:
  • LANGUAGE_AUTO -> auto detect LANGUAGE_NUMERIC, LANGUAGE_DATETIME, LANGUAGE_TEXT
  • TABULAR_AUTO -> "highly-unique categorical columns" (categorical columns that has >100 rows and more than 5% of rows contain unique values) will be detected as TABULAR_CHARACTER
  • AUTO -> "highly-unique categorical columns" will be detected as TABULAR_CHARACTER when lengths of all rows are the same. Otherwise it will be LANGUAGE_TEXT
• Do auto detection before creating a generator (if columns in the config are undefined at all or have one of the *_AUTO encoding types)
• Skip validation of *_model_configuration in SourceTableConfig before auto detection
• Revalidate SourceTableConfig strictly (i.e., including *_model_configuration) after auto detection is done
• Ensure that SourceTable always has strict validation
• New unit tests: test_auto_detect_encoding_types_and_pk, test__auto_detect_encoding_type, test__auto_detect_primary_key

Code refactoring:

• Move main logic for detecting location schema from Core to SDK
• Move logics for auto detection from a isolated script into DataTable class

Misc / Changes that are not directly related to the requirements:

• Update CustomBaseModel in mostlyai/sdk/_data/metadata_objects.py so that the classes can be initialized without using aliases/camelCase field names.
• Fix a bug in MostlyAI.train(): allow data to be fed as the first unnamed argument
• Fix one incorrect test case in test_execution_plan.py
• Add .DS_Store to .gitignore