[YSQL] Connection scaling

[m-iancu]

Jira Link: DB-978
In YSQL, like in vanilla Postgres, each backend connection has some overhead, including to load and maintain the metadata cache for the system catalog tables.
The relative overhead is increased on low-resource machines and geo-distributed cluster when the
This issue tracks tasks to minimize the per-connection overhead for both initializing and maintaining backend connections.

Staging into commits (Release target: v2.16)

Status	Feature	Github Issue	Comments
✅	Functional Spec and Design Doc		TBD -- migrate to github
✅	Load pg_depend and pg_shdepend on demand	#10433	Reduce RPC requests on connection init time
✅	Improve batching for catalog cache (re)load	#10518	Reduce number of per-connection RPCs
✅	Remove YBClient from YSQL/pggate	#9936	Add custom PgClient and PgClientService instead
✅	Add tserver-side cache for syscatalog	#10821	Connections can (re) load their cache with mostly local requests
🕓	Reduce cache-refresh per-connection memory utilization	#14666	Connection cache-refresh over-uses memory as analyzed in YSQL connection memory utilization breakdown
🕓	Pre-load all common catalog tables into metadata cache	#14667	We need to pre-load all commonly used catalog tables to avoid ad-hoc metadata loading during query planning/execution