Split up sql dialects by consuming store

[rbev]

This pull request refactors the SQL dialect and ingestion infrastructure for both PostgreSQL and SQL Server providers, separating the logic for failed message ingestion and retry batch persistence into distinct, provider-specific classes. This change clarifies responsibilities, improves maintainability, and enables more granular testing. It also introduces shared base classes for common SQL generation and execution logic.

Functional changes:

• Improved chunking logic in SQL Server dialects to dynamically calculate maximum rows per statement based on SQL parameter limits. [1] [2] [3]

Refactoring of SQL dialects and ingestion infrastructure:

• Introduced provider-specific base classes (PostgreSqlDialect, SqlServerDialect) to encapsulate common SQL parameterization and execution logic, reducing code duplication. [1] [2]
• Split the ingestion SQL dialect interfaces and implementations:
  • Created IFailedMessageIngestionSqlDialect and IRetryBatchSqlDialect interfaces, with separate implementations for PostgreSQL (PostgreSqlFailedMessageIngestionSqlDialect, PostgreSqlRetryBatchSqlDialect) and SQL Server (SqlServerFailedMessageIngestionSqlDialect, SqlServerRetryBatchSqlDialect). [1] [2] [3]
  • Updated dependency registrations to use the new interfaces and classes. [1] [2]
• Removed duplicate code for parameter row generation and SQL execution from the dialect classes, centralizing it in the new base classes. [1] [2] [3] [4]

Documentation updates:

• Updated docs/error-ingestion-design.md to reflect the new separation of dialect responsibilities, clarifying which classes are responsible for which persistence operations. [1] [2] [3]