Add data source integrations with Snowflake

[sfc-gh-dhung]

Add Snowflake data source integration

This PR adds native support for loading training data directly from Snowflake, enabling users to train models on data stored in their Snowflake data warehouse.

Features

Unified snowflake data source type with three mutually exclusive modes:

• SQL Query (sql): Execute arbitrary SQL queries against Snowflake
• Table (table_name): Load data directly from a Snowflake table (auto-generates SELECT * FROM)
• Dataset (dataset_uri): Load from versioned Snowflake Datasets using the snow:// URI format

Additional options:

• column_mapping: Rename columns from source to target format
• limit: Cap the number of rows loaded
• batch_size: Configure batch size for data retrieval

Configuration

Snowflake credentials can be provided via:

• Environment variables (SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, etc.)
• Snowflake connections config file (~/.snowflake/connections.toml)

Installation

pip install 'arctic_training[snowflake]'

Examples

data:
  sources:
    # From SQL
    - type: snowflake
      sql: "SELECT TEXT FROM ARCTIC_TRAINING.CAUSAL_DEMO.GUTENBERG_100"
      column_mapping: {"TEXT": "text"}
      # limit: 50
      # batch_size: 1024
    # From Snowflake Table
    - type: snowflake
      table_name: ARCTIC_TRAINING.CAUSAL_DEMO.GUTENBERG_100
      column_mapping: {"TEXT": "text"}
    # From Snowflake Dataset
    - type: snowflake
      dataset_uri: "snow://dataset/ARCTIC_TRAINING.CAUSAL_DEMO.GUTENBERG_DATASET/versions/v1"
      column_mapping: {"TEXT": "text"}

Examples are also provided in projects/causal_snowflake