capture-time schema filtering (--schema flag and [capture] config)

[nandorholozsnyak]

Problem

All capture commands (dump-schema, init, snapshot take, snapshot activity) introspect every user schema in the database. There's no way to restrict capture to specific schemas.

Our use case: we have a schema with no PII and a public schema containing personal data (emails, names). We want to use dryrun's offline MCP capabilities but can't commit the snapshot to our repo because it includes public schema objects.

Post-load filtering exists in MCP tools (schema parameter on list_tables, describe_table, etc.) and the lint --schema flag, but PII has already hit disk at that point.

Proposal

Include-list filtering at capture time, configurable via CLI flag and dryrun.toml:

dryrun dump-schema --schema inventory
dryrun init --schema inventory
dryrun snapshot take --schema inventory

[capture]
schemas = ["inventory"]

• CLI --schema (comma-separated) overrides config
• Empty/absent list = capture all (backward compatible)
• Filtering applied after introspection, before any disk write (JSON, history.db)
• Must filter all three streams: DDL, planner stats, activity stats

Notes

• SQL queries in internal/schema/sql/ use //go:embed through QueryStore — Go-level post-query filtering before serialization is lower churn than parameterizing every SQL file, and achieves the same PII guarantee
• search_path in the connection URL doesn't help — introspection queries hit pg_catalog directly, bypassing search_path
• Cross-schema FKs: FK metadata on included tables should be kept even if the referenced table's schema is excluded (the constraint text is string-based)
• Non-existent schema in filter: not an error, just zero objects matched — a stderr warning would be helpful