fix: shut down gateway and metrics server concurrently

[Brunotlps]

Summary

• A single 30s context.WithTimeout was shared between two sequential Shutdown calls
• context.WithTimeout starts its timer at creation time — if srv.Shutdown consumed 29s of the budget, metricsSrv.Shutdown received a nearly-expired context
• This caused active connections (e.g. in-flight Prometheus scrapes) to be aborted instead of drained, with the process exiting 0 as if nothing was wrong
• Both servers now shut down concurrently via goroutines coordinated by sync.WaitGroup, each receiving the full 30s budget

Test plan

• go build ./... succeeds
• go test -race ./... passes with no failures

Closes #7

🤖 Generated with Claude Code