[DPE-3777] Add node port support

[phvalguima]

The node port is created if at least one single unit requests it using external-node-connectivity=True via database relation. Only applications that submit that flag as True receive the nodeport information. Others will receive the internal ClusterIP.

The charm internally checks at update_connection_info() if a given relation needs the NodePort, and updates the service accordingly with patch_port(True).

Periodically, the charm reviews if node port is still necessary. runs: patch_port with external_connectivity output. If all relations are either going away or do not have external-node-connectivity=True, then node port is revoked. This logic is called on all main events changing the state of the charm.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 71.90%. Comparing base (0c2a151) to head (bd2db0a).
Report is 8 commits behind head on main.

❗ Current head bd2db0a differs from pull request most recent head 7f0e0ec. Consider uploading reports for the commit 7f0e0ec to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #264      +/-   ##
==========================================
- Coverage   72.24%   71.90%   -0.35%     
==========================================
  Files           8        8              
  Lines        1081     1075       -6     
  Branches      185      182       -3     
==========================================
- Hits          781      773       -8     
- Misses        230      235       +5     
+ Partials       70       67       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[taurus-forever]

LGTM, but up to @dragomirp here!

[taurus-forever]

As discussed in private, IMHO: we must test 2-3 units here. As in production we have 2+ PGB units for HA.

I believe each client app have to deploy independent mysql-router (for VMs for sure... but I would keep this requirements for K8s as well). Example: pgb normally should be deployen in app mode, so wordpress in model A will have pgb in the same model, but data-integrator will in the DB model OR separate (but outside of model A => different PGB in the real deployment).

Sure, it theory the same PGB should work for 2+ apps... but create a separate bugreport if it is broken here... and we should re-check legacy interface behavior here (app1:psql + app2:postgresql_client => BOOM?).