feat: prometheus histogram for deduct latency

[iexwr]

Summary

Adds a dedicated Prometheus histogram for end-to-end latency tracking of POST /api/billing/deduct.

Changes

• src/metrics/registry.ts — New file registering billing_deduct_duration_seconds histogram with explicit buckets from 1ms to 10s and route + status_code labels
• src/middleware/metricsHistogram.ts — New per-route middleware that measures request latency using performance.now() and records to the histogram on response finish
• src/routes/billing.ts — Wired billingDeductHistogramMiddleware into the /deduct route handler
• docs/grafana-dashboard-billing-deduct.json — New Grafana dashboard with P50/P95/P99 latency percentiles and cumulative distribution panels
• src/tests/billingDeductMetrics.test.ts — 21 unit tests covering histogram registration, bucket boundaries, label correctness, middleware behavior, error status codes, reset, and dashboard consistency

Acceptance Criteria

• Histogram exported (billing_deduct_duration_seconds)
• Buckets cover 1ms..10s (12 explicit buckets)
• Route label present (/api/billing/deduct)
• Dashboard JSON updated
• Tests pass (21 new tests, all 79 metrics tests passing)
• Lint clean (0 errors)

Test Output

PASS src/__tests__/billingDeductMetrics.test.ts
Tests:       21 passed, 21 total

PASS src/__tests__/metricsLatency.test.ts
Tests:       43 passed, 43 total

PASS src/__tests__/upstreamProfiling.test.ts
Tests:       15 passed, 15 total