fix(db): tolerate a SQLite build without the dbstat virtual table

[TechNickAI]

The bug

getDatabaseStats() queried the dbstat virtual table once per table with no guard:

const tableSize = db
  .prepare(`SELECT SUM(pgsize) as size FROM dbstat WHERE name = ?`)
  .get(table.name) as { size: number | null };

dbstat is compile-time optional in SQLite (ENABLE_DBSTAT_VTAB). It is not a schema table, is never created by a migration, and does not appear in sqlite_master. Builds without it — sql.js/WASM among them — reject the query outright.

openSqliteDatabase() (src/lib/db/core.ts:178) falls back to a sql.js WASM adapter when no sync native driver is available, so this is reachable in normal operation.

Impact

The throw propagates out of getDatabaseStats() and takes every caller with it. Most visibly, GET and PATCH /api/settings/database return HTTP 500:

Error getting database settings: Error: no such table: dbstat
    at src/lib/db/stats.ts

The entire database settings page becomes unusable — page size, cache size, and vacuum settings can be neither read nor changed. Observed in production on a live router.

The fix

The function already anticipated missing virtual-table modules: the COUNT(*) lookup a few lines above swallows "no such module:" errors. The dbstat query simply sat outside that guard.

Probe dbstat once per call and skip the per-table size lookups when unavailable, reporting size: 0. Database-level figures (total size, page count, cache size) come from pragmas and stay accurate — only per-table byte sizes are lost, which is the right trade against a hard 500.

Unrelated failures (I/O errors, corruption) still propagate.

Both spellings are handled, and the match is deliberately unanchored because drivers prefix their error class onto the message (SqliteError: no such table: dbstat, RuntimeError: ...). A test pins this so the guard cannot silently regress if anchored later.

Tests

tests/unit/db/stats-dbstat-optional.test.ts — 8 cases against a SqliteAdapter double:

• dbstat available → real per-table sizes
• no such module: dbstat → degrades to 0
• no such table: dbstat → degrades to 0
• prefixed (SqliteError: / RuntimeError:) → degrades to 0
• database with no user tables
• SUM(pgsize) returning NULL → 0
• dbstat failing after a successful probe → propagates (a mid-iteration fault is a real fault, not an absent module)
• database disk image is malformed → propagates

Falsified, not assumed: reverting the guard makes exactly 2 of these fail with the pre-fix 500 path. Introducing a ^ anchor makes exactly the prefixed-error case fail.

Verification

tests/unit/db/**       110 pass, 0 fail
tests/unit/api/**      215 pass, 0 fail
npm run lint           PASS (whole repo)
typecheck:core         0 errors
prettier --check       clean
check:db-rules         PASS
check:cycles           PASS
check:file-size        PASS
check:test-discovery   PASS
check:any-budget:t11   PASS

Notes for review

• Behaviour change: callers now see size: 0 for every table on a dbstat-less build instead of an exception. No caller was found that depends on size being truthy.
• The probe is per-getDatabaseStats() call rather than cached on the adapter. Caching capability at connection setup would be tidier; that felt like a larger change than this fix warrants, and I'd happily follow up if preferred.
• Based on 2fc1229fe (v3.8.50), the commit currently deployed in our production environment.