test(auto): seed eligible accounts in chaos combo fixture

[diegosouzapw]

After #15979, provider-wide ToS filtering can legitimately remove anonymous catalog candidates. The existing auto/chaos test created no accounts and therefore received an empty pool while expecting a populated chaos panel.

Set temporary data/plugin directories before importing the factory, and seed eligible OpenAI and Anthropic accounts through the DB module in beforeAll. Close the DB and restore the environment after the suite. All four existing test bodies and assertions remain byte-for-byte unchanged.

Validation on the pinned base f90e64a1c9921b91f40c63e039fb4521f28fa8bc:

• Initial local reproduction, using the copied dependency environment: 3 passed / 1 failed at config.chaos.enabled before the fixture change.
• After a normal npm ci with the exact lockfile: all 4 chaos tests and all 8 existing catalog/ToS cases passed, covering default exclusion, explicit opt-out and per-model precedence.
• Core/dashboard type checks, targeted ESLint/Prettier, file-size, complexity/cognitive ratchets, discovery, mutation coverage, test-masking and changelog checks passed with the exact dependencies.
• Full ESLint passed in four disjoint CLI groups using unchanged rules/suppressions and fresh caches. The actual ESLint 10.12 selector and final re-enumeration verified all 12,631 files exactly once: zero errors, warnings or fatal results; all four exits were zero.
• The canonical quality:gate passed using the complete, verified report through its supported ESLINT_RESULTS_JSON input (430.68 seconds).
• Normal commit hooks passed. Independent review of the identical one-file patch found no issues and independently reran the four chaos tests.

Refs #15979 and #15059. This follow-up covers the test fixture; the dashboard opt-in/ToS notice requested in #15059 remains outside its scope.

[github-actions]

CI Coverage Report

• Coverage job: skipped
• PR test policy: success

Coverage artifact was not available for this run.