Handle JsonRpcService startup failure

[jbiseda]

Problem

In JsonRpcService::new if start_http fails close_handle_sender will be dropped without sending a value to the channel. Subsequent close_handle_receiver.recv().unwrap() will fail on the closed channel.

Summary of Changes

Send error value over the channel to gracefully handle the failure. Handle errors gracefully in Validator::new rather than using validator::abort().

See #25400 for original change which was reverted by #26130 due to #25941.

Description of issue with original change from #25941:

Problem seems to be that Validator::new fails using abort() which in tests calls panic(). This is handled by the test framework and can leave rocksdb background threads accessing partially torn down state. This is exacerbated by tests in program.rs and commands.rs running parallel tests which call Validator::new. This can be mitigated by running tests serially and using rocksdb DB::cancel_all_background_work.

From a cursory examination of rocksdb code it looks like background threads are cleaned up when the DB object is torn down. One solution is to have Validator::new return a Result and fail without using abort() which will tear down state cleanly.

Fixes #

[lijunwangs]

Looks good to me