File query: filtering (WHERE clause via DuckDB)

## Context

Part of the DuckDB file query engine epic (#147). Depends on #148 (FileQueryEngine) and #149 (query editor UI). The original implementation plan proposed a custom parser and Rayon evaluator — with DuckDB this is replaced entirely by standard SQL `WHERE`.

## What this covers

Exposing DuckDB's `WHERE` clause through the query editor for structured filtering of open files.

### Examples (typed into the query editor)

```sql
SELECT * FROM data WHERE response_time > 500
SELECT * FROM data WHERE status_code != 200 AND host = 'prod'
SELECT * FROM data WHERE price < 100 AND stock > 0
SELECT * FROM data WHERE timestamp > '2024-01-01' AND error_level = 'critical'
SELECT * FROM data WHERE user.age >= 18
SELECT * FROM data WHERE deleted_at IS NULL
```

DuckDB handles type inference, nested field access via dot notation (with `json_extract` under the hood for JSON), NULL semantics, operator precedence, and parenthesised grouping — no custom parser needed.

## What needs building

This ticket is **UI/UX only** — the engine work is in #148, the editor widget in #149. The work here is:

- Validate that DuckDB's JSON path handling covers nested fields from plugin-loaded JSON records (e.g. `user.age`, `items[0].price`)
- Handle the case where `data` schema has dynamic / heterogeneous columns (wide sparse tables from heterogeneous JSON) — surface schema inference errors clearly
- Test filtering across all three ingestion paths: native JSON, native CSV, plugin-ingested

## Acceptance criteria

- [ ] `WHERE` with comparison operators (`>`, `<`, `>=`, `<=`, `!=`, `=`) works on numeric and string columns
- [ ] Logical operators (`AND`, `OR`) with parentheses work correctly
- [ ] Nested field access (`user.age`, `items.length`) works for JSON sources
- [ ] `IS NULL` / `IS NOT NULL` work
- [ ] Type mismatch errors shown inline in the query editor (not a crash)
- [ ] Works for native JSON, native CSV, and plugin-ingested formats