Graceful shutdown blocked by Keep-Alive connections

The idea behind graceful shutdown is that when shutting down graphql-engine, 1. it stops accepting requests, 2. it handles in-flight requests 3. it exits.

Due to a bug in the underlying web server https://github.com/yesodweb/wai/issues/853, in step 1 instead of stopping to accept requests, it only stops accepting new connections, but keeps accepting requests on any open keep-alive HTTP connections, until shutting down forcibly after the graceful shutdown timeout of 30s.

(Essentially, this is a duplicate of #2698, we might opt to reopen that ticket instead. Not sure how that should be handled.)

### Version Information

Server Version: current HEAD

### Environment

any

### What is the expected behaviour?

- If graphql-engine is interrupted while there are idle connections but no in-flight requests, it should exit more or less immediately, not wait for the graceful shutdown timeout of 30s.
- After graphql-engine is interrupted, it should not accept requests on existing HTTP connections.

### Keywords

Keep-Alive, HTTP, shutdown

### Any possible solutions?

This should ideally be fixed upstream in Warp, and then we should upgrade to that version.