[datlechin]

Summary

• Match DuckDB 2.0 quoting and SQL Server statement starts when splitting SQL and checking writes.
• Resolve PostgreSQL catalog names through built-ins while preserving the session's schema and table selection.
• Keep read connections read-only, omit more credential SQL from history, and bound database reply parsing. Read nested MongoDB documents without exhausting worker stacks or copying each subtree; report count decoding errors.
• Show import renames, SSH servers and TLS changes. Keep unchanged names and recover older group and tag names without merging their connections.

Test plan

• Add SQL splitting, classification, history, driver read, parsing-limit and import roundtrip regressions.
• Update existing DuckDB and PostgreSQL lexer expectations to match the engines; keep the shared function fixture unchanged from main.
• Add PostgreSQL tests for empty paths, catalog-first paths, table-name collisions and failed lookup recovery.
• Local setup and build: PASS with lld and one job. Full suites: FAIL on MongoDB nesting; the corrected decoder and remaining checks await verification.
• Retain the 128-level BSON regression on a 2 MiB stack and exercise the production decoder; add container-shape, large-value, malformed-data and field-order cases.