Stop wiping the whole database between tests

[fajarhide]

startTestServer ran TRUNCATE workspaces CASCADE on every call, so two suite runs against one TEST_DATABASE_URL deleted each other's rows mid-test. The failures came out as expected 401 to be 200, because the credential row went away between the seed and the request, and nothing in the output suggested another process was involved.

Nothing needed the wipe. Every test already creates its own workspace, and credentials, grants, enablements and usage rows all cascade from it. One assertion did read the whole table, counting oauth_states to prove an expired row was swept; it now names the state it is about.

Before and after, both suites started at the same moment against the same database.

On main:

$ npx vitest run test/mcp.test.ts test/rest.test.ts
Tests  6 failed | 9 passed (15)

$ npx vitest run test/connections-http.test.ts test/isolation.test.ts test/admin-connections.test.ts
Tests  7 failed | 13 passed (20)

On this branch, same two commands, same overlap:

Tests  15 passed (15)
Tests  20 passed (20)

Full suite: 35 files, 351 tests, green. Typecheck clean.

Worth naming the tradeoff: rows now accumulate in the test database across runs instead of being cleared. Nothing reads across workspaces, so that is invisible to assertions, and a dropdb is the reset if it ever gets large.

Closes #15