Add DatabaseDialect and implement identifiers truncation

[martinv13]

Two related problems motivated this refactor:

• Backend-specific logic was scattered across 6 files as if db_type == "..." conditionals, making it hard to maintain.
• PostgreSQL limits identifiers to 63 characters, which caused errors with longer XSD element names.

Changes:

• Added a DatabaseDialect class hierarchy (base, postgresql, mssql, mysql, duckdb), one subclass per backend.
• Migrated all db_type conditionals into the appropriate dialect class.
• db_identifier() in the base class now truncates names exceeding MAX_IDENTIFIER_LENGTH using a stable MD5 hash suffix. Each dialect sets its own limit.
• SQLAlchemy's key/name duality on Column is used to keep logical names stable in Python while sending truncated physical names to the database.