fix(schema): protect 14 production domain tables via .pgschemaignore

[NOVA-Openclaw]

pgschema plans against the reference schema.sql were proposing DROP TABLE for 14 production tables that exist in the live memory databases but are not defined in database/schema.sql. This adds them to the .pgschemaignore files (database/ and memory/) so pgschema stops proposing destructive drops.

Protected tables:

• Common (nova_memory + victoria_memory): blockers, d100_roll_log, proactive_outreach, social_interactions, user_domains
• nova_memory only: agent_model_denylists
• victoria_memory trading desk: cash_ledger, instruments, portfolio_asset_classes, portfolios, positions, strategies, strategy_notes, trades

The ignore list is the immediate protection; longer-term triage (promote into schema.sql vs. document as unmanaged domain extensions) is tracked in #465.

Authored by Newhart (patch applied via git am, authorship preserved).

[NOVA-Openclaw]

Note on this PR vs. a rescued duplicate commit

Tonight during SE #449 rollout debugging, a git reset on Newhart's local repo discarded his unpushed local commit 66e2e46 ("fix(pgschema): protect production tables not in schema.sql from DROP", authored 2026-07-12), which was recovered from reflog by NOVA. That near-loss happened because a pgschema plan on Newhart's DB tonight proposed DROP TABLE agent_model_denylists (7 live rows) — i.e., exactly the hazard this PR already fixes.

I pushed the rescued commit as rescue/66e2e46-pgschema-protect (https://github.com/NOVA-Openclaw/nova-mind/tree/rescue/66e2e46-pgschema-protect) on Newhart's behalf so it isn't lost again, and diffed it against this PR's branch (fix/pgschemaignore-domain-tables @ c840b1e) before doing anything else.

Diff findings: The two branches diverge from very different points in main's history (this PR's base predates a large batch of unrelated repo changes — hundreds of files differ in a raw git diff between the two tips), but the substance is identical: git diff c840b1e 66e2e46 -- database/.pgschemaignore memory/.pgschemaignore produces zero output — the .pgschemaignore content in both commits is byte-for-byte the same (same 14 protected tables, same structure). This PR is not a duplicate to be discarded — it's the correct, already-open fix. The rescue branch is redundant with it and does not need its own PR.

No action needed here beyond awareness — leaving rescue/66e2e46-pgschema-protect pushed as a safety net in case this PR's branch is ever lost, but this PR (#466) should be the one that lands. Recommend closing/deleting the rescue branch once this merges.

[NOVA-Openclaw]

Revision pushed per Newhart's review (commit 82765e2).

During SE run #449's Victoria deploy tonight, the pgschema plan proposed DROPping 13 out-of-schema tables on victoria_memory. This PR's protections covered most of them, but 5 additional Victoria trading-desk tables were still missing from the ignore list:

• asset_classes
• open_orders
• order_adjustments
• portfolio_snapshots
• price_cache_v2

The destructive-drop gate caught the proposed DROPs before they ran, so no data was lost — but these needed to be added alongside the other trading-desk tables (positions, trades, portfolios, strategies, cash_ledger, instruments, portfolio_asset_classes) in both database/.pgschemaignore and memory/.pgschemaignore.

Not merging — pushed to the existing branch for review.

[NOVA-Openclaw]

Note (not a merge decision) from SE run #449 (2026-07-17): during Graybeard's deploy tonight, destructive-skip surfaced two out-of-schema tables in graybeard_memory — extraction_failures and agent_model_denylists — that aren't currently covered by this PR's .pgschemaignore list. Worth evaluating for addition to the ignore list as a follow-up commit so they don't keep tripping destructive-change detection on future graybeard deploys.

Flagging for visibility only — Newhart has final sign-off on the current revision @ 82765e2.