Feat: finalize clickhouse engine adapter

[treysp]

This PR finalizes the Clickhouse engine adapter:

• Adds SCD model kind support
• Adds ability to pass arbitrary settings to client connection
• Adds docs

Context and implementation details

Joins and NULLs

• Clickhouse defaults to filling empty cells with a datatype-specific default (e.g., 0 for integer columns).
• The SCD and table diff queries SQLMesh builds require that we change that behavior to fill with NULLs.
• We do that by injecting SETTINGS join_use_nulls = 1 into the query
• SCD detail:
  • The original user query is embedded in a CTE.
  • Query settings are dynamically scoped, so our setting on the outer query will apply to the user query CTE.
  • If join_use_nulls = 0 on the CH server, we inject join_use_nulls = 0 into the CTE to preserve behavior expected by user.

Connection settings

• Following dbt-clickhouse (maintained by Clickhouse), we pass these settings to the connection:
  • Always
    • mutations_sync = "2"
    • insert_distributed_sync" = "1"
  • When running in cluster or cloud modes
    • database_replicated_enforce_synchronous_settings = "1"
    • insert_quorum = "auto"

Storage format

• Each CH table must have a "table engine" (MergeTree by default)
• Users specify a table engine in the MODEL DDL storage_format key
• Specification may be a function call, so we now generate the value's SQL if the value is an Expression other than Literal or Identifier
  • Generated Literal/Identifier SQL may include quotes, and we want to defer normalization until later

[erindru]

Awesome work! I bet you learned way more than you wanted to about the Clickhouse internals

[georgesittas]

Great work @treysp!