feat(mental-models): expose refresh operation model id and list total

[benfrank241]

Two small observability improvements around mental models. Independent of each other; no schema changes.

1. mental_model_id on operation records

refresh_mental_model operations return document_id: null and no model identifier, so a client can't tell which mental model an operation refreshed (diagnosing refresh behaviour from the operations log is impossible). The id already exists in the operation row's result_metadata — this surfaces it as a first-class mental_model_id field on OperationResponse and OperationStatusResponse (null for other task types).

2. total on GET /banks/{bank}/mental-models

The list envelope was {"items": [...]} only and silently capped at limit (default 100), so a bank with more models than the page size looked fully-listed. Adds total (matching-row count, ignoring limit/offset) plus limit/offset echoes, backed by a new count_mental_models() engine method that applies the same tag filter as list_mental_models.

Tests

• A refresh_mental_model operation surfaces the correct mental_model_id; non-refresh ops don't carry it.
• Listing with a smaller limit still reports the full total, including a tag-filtered total.

🤖 Generated with Claude Code

[koriyoshi2041]

The API shape and focused tests look consistent, but this head cannot merge yet because the checked-in generated surface is stale. verify-generated-files regenerated 10 files (347 insertions / 9 deletions), including the Go/Python/TypeScript response models plus both OpenAPI copies. Please run the generation commands printed by the job and commit those outputs.

The separate API shard failure looks unrelated to this diff: 1,940 tests passed and the only failure was a PostgreSQL deadlock in test_import_bank_creates_per_bank_indexes. I would treat that as a rerun candidate after the generated files are updated.

[nicoloboschi]

superseded by #3548 including same fix for directives