[tonyandrewmeyer]

A collection of small model.py fixes, none of which seemed worth a PR of its own:

• A charm subclassing a built-in status (to add a field, say) replaced the built-in in StatusBase._statuses, so from_name() constructed the subclass when reading the status back from Juju and raised TypeError on every read. Built-in names now keep their built-in class.
• RelationData held only a weakref.proxy to its Relation (since the original 2019 code, with no recorded reason). A Relation that the model doesn't cache (one fetched by an ID that relation-ids doesn't list, as happens during relation-broken) was collected as soon as the caller dropped it, and then every access through .data raised ReferenceError.
• Secret kept a reference to the caller's content dict rather than a copy.
• Network parsing used strict ip_network(), which raises ValueError for a CIDR with host bits set, such as the interface's own address with the subnet's prefix length.
• LazyNotice and LazyCheckInfo recursed forever in __getattr__ when copied, because copy looks the private attributes up before __init__ has run.
• set_ports compared the protocol case-sensitively against opened-ports, so an upper-case protocol closed and reopened the port. open_port and close_port already lowercase it, so this brings the three into line (the type hints forbid upper case anyway).
• An unknown secret rotation policy from secret-info-get raised ValueError.
• The get_content docstring claimed that set_content invalidates the cached content, which the code deliberately doesn't do (and the Scenario tests assert that it doesn't). Juju's hook context returns a pending secret-set value for peek or refresh in the same hook and the tracked revision otherwise, so the docstring now says that.