test: add metering E2E integration tests + Ollama CI tier

[pyramation]

Summary

Test #3 from constructive-planning#911: full metering pipeline E2E tests + new Ollama CI tier.

Metering E2E tests (metering-e2e.test.ts, 15 tests)

Config resolution (3 tests):

• getLlmBillingConfig resolves billing + inference log config from metaschema tables
• Returns null for unknown database_id

Billing functions (4 tests):

• check_billing_quota returns true when unlimited (no credits row)
• Returns true when under limit, false when over limit
• record_usage writes to balance + ledger with metadata

Inference log (1 test):

• logInferenceUsage inserts row with all fields (service, operation, cache tokens, etc.)

meteredEmbed with mock embedder (4 tests):

• Calls embedder + records usage when quota allows
• Returns quotaExceeded when limit reached
• Passes through unmetered when no meter slug configured
• request_id propagates to billing ledger metadata

meteredEmbed with real Ollama (3 tests, skipped when unavailable):

• Full pipeline: Ollama embed → billing check → record_usage → inference_log + ledger
• Quota enforcement blocks real Ollama calls
• Multiple calls accumulate in balance and ledger

CI Tier 4 — Ollama tests (run-tests.yaml)

New CI job with PostgreSQL + Ollama service containers:

• Uses bash /dev/tcp health check (Ollama image doesn't have curl)
• Pulls nomic-embed-text model before test run
• Runs only metering-e2e tests (existing graphile-llm.test.ts tests require schema setup from pg-tests tier)
• Tests gracefully skip Ollama-specific cases when Ollama is unavailable (local dev)

Test infrastructure (metering-setup.sql)

Minimal stub schemas matching what config-cache.ts queries:

• metaschema_modules_public.billing_module + inference_log_module
• Simplified but functional check_billing_quota + record_usage functions
• usage_public.usage_log_inferences table
• Seeded meter rows for embedding + inference

Review & Testing Checklist for Human

• Verify the Ollama CI tier doesn't slow down the critical path (it runs in parallel with other tiers)
• Check that metering-setup.sql stub functions match the real billing function signatures
• Confirm nomic-embed-text is a reasonable model choice for CI (small, fast)

Notes

• Companion PR for Test readme #2 (inference rollup tests): constructive-db#1292 (merged)
• The 3 Ollama tests are describe.skip'd when SKIP_OLLAMA_TESTS=1 or Ollama is unreachable — safe for local dev
• The setup SQL uses CREATE FUNCTION (not CREATE OR REPLACE) — these are test-only stubs in ephemeral databases

Link to Devin session: https://app.devin.ai/sessions/2b5a29d83d3f478e8d3d972653b4879c
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[devin-ai-integration]

Closing — metering E2E tests should live in constructive-db where the real billing infrastructure exists, not stub SQL in the open source repo.