chore(async): track fire-and-forget goroutines, drain on shutdown

[lakhansamani]

Why

Detached go func() calls on the request path (email/SMS send, webhook
events, audit log writes) were invisible to graceful shutdown and, if they
panicked, took down the entire process — not just the one request.

What

• New internal/asyncutil package: Go(log, fn) tracks the goroutine and
  recovers any panic; Wait(log) blocks until all tracked goroutines finish.
• Every fire-and-forget goroutine in internal/service/, internal/http_handlers/,
  and internal/audit/ now goes through asyncutil.Go instead of a bare go func().
• internal/server/server.go calls asyncutil.Wait after the HTTP/metrics
  listeners stop, before the process exits.
• Long-running loops already tied to ctx (listeners, ticker-based cache
  eviction) are untouched — they don't need draining.

Test plan

• go build ./..., go vet ./...
• internal/asyncutil unit tests (drain + panic-recovery)
• TEST_DBS=sqlite go test -p 1 ./internal/... — full suite green