[TRACKING ISSUE] Refactor client connection manager to simplify

The tracking issue for the Java side PR.

See https://github.com/hazelcast/hazelcast/pull/20806 for details.

---

Most of the logic is turned into a single threaded one to
be able reason about the logic easily.
Important ones that is responsibility of the single thread
- Writing to and removing from activeConnections
- Checking if all connections are gone and initating new cluster
- Checking if cluster id has changed and initating failover
if necessary.

We still connect to other members in parallel because we mainly
want a fast start. We gave up on one thing. We no longer try to
open connection to all members while trying to connection back
to a cluster. This does not effect initial start since in the
initial start, we had to wait for initial connection and member-list
from it to be able to connect all members anyway.