Add document change event broadcast using members list

[dc7303]

What this PR does / why we need it:
It broadcasts document change events to replicated agents.

• Add grpc broadcast service.
• Implement the Publish API.
• Broadcast client management.
• Add grpc_recover.
• When requesting from the server to the server, the server should not be stopped when the replicated agent is in a state where communication is not possible.
• Add stress test
• Add design document for broadcast
• Code cleanup

Added issue

• If you enable the server in multiple tests for the cluster mode test, they affect each other and fail.
  example)

 Error Trace:    cluster_stress_test.go:32
                Error:          Received unexpected error:
                                listen tcp :21501: bind: address already in use
---
 Error Trace:    cluster_mode_test.go:59
                                cluster_mode_test.go:128
                Error:          "map[c2oi......]" should have 2 item(s), but has 7

Which issue(s) this PR fixes:

Address #183

Special notes for your reviewer:

Does this PR introduce a user-facing change?:

Additional documentation:

Checklist:

• Added relevant tests or not required
• Didn't break anything

[dc7303]

yorkie/test/helper/helper.go

Lines 110 to 136 in a1b8407

	var portOffset = 0
	// TestConfig returns config for creating Yorkie instance.
	func TestConfig(authWebhook string) *yorkie.Config {
	portOffset += 100
	return &yorkie.Config{
	RPC: &rpc.Config{
	Port: RPCPort + portOffset,
	},
	Metrics: &prometheus.Config{
	Port: MetricsPort + portOffset,
	},
	Backend: &backend.Config{
	SnapshotThreshold: SnapshotThreshold,
	AuthorizationWebhookURL: authWebhook,
	},
	Mongo: &mongo.Config{
	ConnectionURI: MongoConnectionURI,
	ConnectionTimeoutSec: MongoConnectionTimeoutSec,
	PingTimeoutSec: MongoPingTimeoutSec,
	YorkieDatabase: TestDBName(),
	},
	ETCD: &etcd.Config{
	Endpoints: ETCDEndpoints,
	},
	}
	}

Port conflicts occur when multiple packages use this function because it recompiles when the package being tested is replaced. I am looking for a way to use ports stably to run multiple servers.
go-test doc

'Go test' recompiles each package along with any files with names matching
the file pattern "*_test.go"
....

[hackerwins]

Thanks for your contribution.
I left a few simple questions.

[hackerwins]

Is there any reason we broke the error chain here?

[dc7303]

I didn't think it was necessary to return an error from the API where a particular Document event occurred just because it couldn't broadcast to another agent.
So I left an error log at the time of occurrence. This is because we thought that document sync could be synced through other sync events.

I have been thinking about how to alert when a problem that can occur within the method (such as a convert failure or an abnormal problem in the target agent) occurs continuously, but I was not sure, so I tried to suggest it later.

Can you please let me know if there are more things I should consider?

[hackerwins]

When we try to convert an event type that is not supported, an error occurs, and there is no need to loop further in the caller. I wish we could handle errors explicitly.

[codecov]

Codecov Report

Merging #189 (28cc71b) into main (b0af224) will decrease coverage by 1.21%.
The diff coverage is 22.61%.

❗ Current head 28cc71b differs from pull request most recent head 6d267ab. Consider uploading reports for the commit 6d267ab to get more accurate results

@@            Coverage Diff             @@
##             main     #189      +/-   ##
==========================================
- Coverage   61.58%   60.36%   -1.22%     
==========================================
  Files          41       42       +1     
  Lines        3454     3550      +96     
==========================================
+ Hits         2127     2143      +16     
- Misses       1131     1211      +80     
  Partials      196      196              

Impacted Files	Coverage Δ
api/converter/from_pb.go	65.22% <0.00%> (-1.88%)	⬇️
api/converter/to_pb.go	82.93% <0.00%> (-2.72%)	⬇️
client/client.go	12.13% <0.00%> (ø)
yorkie/backend/sync/etcd/membermap.go	0.00% <0.00%> (ø)
yorkie/backend/sync/etcd/pubsub.go	0.00% <0.00%> (ø)
yorkie/backend/sync/memory/coordinator.go	0.00% <0.00%> (ø)
yorkie/rpc/cluster.go	0.00% <0.00%> (ø)
yorkie/rpc/server.go	50.72% <3.57%> (+0.91%)	⬆️
yorkie/backend/sync/memory/pubsub.go	91.89% <89.18%> (-1.33%)	⬇️
yorkie/backend/sync/etcd/client.go	71.42% <100.00%> (+1.73%)	⬆️
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b0af224...6d267ab. Read the comment docs.

[hackerwins]

Thank you for your work. In particular, changing the topic to multi-key based in PubSub seems to be neat.

Broadcast seems to have many weaknesses compared to Gossip-based algorithms, but first of all, it seems to be able to operate a small-scale cluster.

(I made a few changes during the review.)