Full import and export: object scope, dialect-correct dumps, native dump tools per engine

[datlechin]

Problem

Import and export cover tables and views only. Everything else a real dump needs is missing:

• No routines, triggers, events, sequences, user types, or grants in an export.
• No per-table WHERE, row limit, or column subset.
• SQL export has no insert mode (IGNORE / REPLACE / UPSERT), no consistent snapshot, no file splitting, no engine-specific header.
• Native dump tools exist for PostgreSQL only (pg_dump / pg_restore). No mysqldump, mongodump, sqlite3 .dump, bcp, ClickHouse native, DuckDB EXPORT DATABASE.
• No XLSX, Parquet, NDJSON, or TSV import. No Parquet, NDJSON, TSV, Markdown, HTML, or COPY export.
• No saved export profiles, no table-to-table transfer between two connections.
• export_data over MCP writes its own formats instead of reusing the export plugins, so there are two sources of truth.

Proposed solution

Five phases, in order:

1. Object scope. Generalize PluginExportTable to an object with a kind (table, view, routine, trigger, event, sequence, type, grant) and add defaulted fetch methods to PluginExportDataSource. Export tree becomes database > object group > object.
2. Data scope and SQL dialect. Per-table WHERE / LIMIT / column subset. SQL options for insert mode, consistent snapshot, hex blobs, complete inserts, split by size or rows.
3. Native dump per engine. Generalize PostgresDumpService into a service driven by a descriptor each driver declares (binary names, arguments, environment, progress source). Three models to keep separate: shell-out (MySQL, MongoDB, SQL Server), in-process SQL (SQLite, DuckDB, ClickHouse), and server-side to object store (BigQuery, Snowflake, Oracle Data Pump).
4. Formats and transfer. Parquet, NDJSON, TSV, Markdown, HTML, COPY on export; XLSX, Parquet, NDJSON, TSV on import. Connect an export data source straight to another connection's import sink for table-to-table transfer with no file.
5. Profiles and parity. Saved export profiles per connection, error report file for skipped rows, Safe Mode gating on row imports, and export_data reusing the export plugins.

Phase 1 changes PluginKit transfer types, so it needs an additive overload and has to ship with an app release, not ahead of one.

Alternatives considered

Adding formats one at a time without opening the object scope first. That leaves every new format stuck exporting tables and views, and the same work has to be redone later.

Related database type

N/A / General