Avoid incoming request hanging when Interactive engine is processing a long-run request

[zhanglei1949]

As being tested, the interactive engine blocks requests when a complex, long-run query is being running.

Observations:

• The http handler could only process the next request only when current request has been finished.
• Seastar decides which shard will handle the incoming request.

[zhanglei1949]

[Failed attempt] Start from this intuition, Is it possible to launch AdminService on some seastar shards, and QueryService on other shards?
However, it is not possible. scylladb/seastar#2183
And this has been confirmed by me with some tests on seastar httpd.

It turns out that if we choose shards to start services, the http request will be lost.

[zhanglei1949]

We could forwarding the request handling to desired shards, in http_handler, which may save us from this issue, although it may waste the computing resources.