Oximeter creates and destroys qorb pools while waiting for schema migration

[jgallagher]

When Oximeter starts up and clickhouse is on an old schema version, it goes into an infinite loop waiting for an operator to run schema migration:

omicron/oximeter/collector/src/lib.rs

Lines 278 to 284 in 2a9f164

	let agent = backoff::retry_notify(
	backoff::retry_policy_internal_service(),
	make_agent,
	log_client_failure,
	)
	.await
	.expect("Expected an infinite retry loop initializing the timeseries database");

However, in every iteration of this loop, we create a new qorb pool and then drop it when returning the error indicating the version isn't what we expect:

omicron/oximeter/collector/src/agent.rs

Lines 704 to 719 in 2a9f164

	let client = Client::new_with_pool(native_resolver, &log);
	match client.check_db_is_at_expected_version().await {
	Ok(_) => {}
	Err(oximeter_db::Error::DatabaseVersionMismatch {
	found: 0,
	..
	}) => {
	debug!(log, "oximeter database does not exist, creating");
	client
	.initialize_db_with_version(
	replicated,
	oximeter_db::OXIMETER_VERSION,
	)
	.await?;
	}
	Err(e) => return Err(Error::from(e)),

In combination with qorb versions prior to the fix in oxidecomputer/qorb#78, this results in 8 leaked TCP connections for every retry attempt.