Migrate zaza from libjuju (async) to jubilant (sync)

[freyes]

Replace the libjuju-based async machinery with jubilant, a synchronous
CLI wrapper around the juju binary. The public API surface is preserved
so that downstream consumers (e.g. zaza-openstack-tests) continue to
work without changes.

• Remove the libjuju background-thread infrastructure: event loop
  management, _libjuju_thread/_libjuju_loop globals, libjuju_thread_run(),
  get_or_create_libjuju_thread(), join_libjuju_thread(),
  clean_up_libjuju_thread(), and run_coroutine_threadsafe().

• Set RUN_LIBJUJU_IN_THREAD = False (no-op constant kept for compat).

• Rewrite sync_wrapper() as a transparent pass-through: if the wrapped
  callable returns a coroutine it is drained with a fresh event loop;
  otherwise the result is returned directly. This allows async functions
  that have not yet been rewritten to keep working without a background
  thread.

• Retain clean_up_libjuju_thread() / join_libjuju_thread() /
  get_or_create_libjuju_thread() as no-op stubs for downstream compat.

• Rewrite all functions using jubilant.Juju() instead of libjuju:

  • add_model(): jubilant.Juju().cli('add-model', ...)
  • destroy_model(): polling loop with jubilant.Juju().cli('destroy-model')
  • list_models(): jubilant.Juju().cli('models', '--format', 'json')
  • cloud() / get_cloud(): jubilant.Juju().cli('clouds', '--format', 'json')

• Remove all async/await keywords; functions are now plain synchronous.

• Fix pre-existing syntax errors: 'model_name=None: Optional[str]'
  corrected to 'model_name: Optional[str] = None' in deployed(),
  get_unit_from_name(), get_model(), scp_to_unit().

• Fix incorrect import: ModelInfo -> ModelStatus from jubilant.

• Add 'import dataclasses' at the top level.

• Add get_status_jubilant(model_name=None):

  • Calls jubilant.Juju(model=model_name).status().
  • Collects subordinate unit statuses from each principal unit's
    .subordinates dict and injects them back into the subordinate
    application's units dict using dataclasses.replace() (required
    because jubilant Status is a frozen dataclass).
  • Returns a fully-populated jubilant Status object.

• Add adapter classes for the wait_for_application_states polling loop:

  • _JubilantUnitAdapter: wraps jubilant UnitStatus; exposes
    workload_status (str), workload_status_message, entity_id, name,
    application, machine (None), data (agent-status dict).
  • _JubilantAppAdapter: wraps jubilant AppStatus; .units returns a
    list of _JubilantUnitAdapter objects.
  • _JubilantModelAdapter: wraps jubilant Status; exposes .units (flat
    dict across all apps) and .applications (dict of _JubilantAppAdapter).

• Add libjuju-compatibility wrappers for callers of get_status():

  • _LibjujuUnitCompat: exposes .get('subordinates') and dict-style
    access to unit fields.
  • _LibjujuAppCompat: exposes .status.status and ['units'] dict.
  • _LibjujuStatusCompat: exposes .applications dict of _LibjujuAppCompat.

• Override get_status() to return _LibjujuStatusCompat(get_status_jubilant())
  so that callers such as failure_report() and get_subordinate_units()
  continue to work unchanged.

• Rewrite wait_for_application_states() as a fully synchronous polling
  loop:

  • Uses time.sleep(1.0) between iterations instead of asyncio.sleep.
  • Calls get_status_jubilant() each iteration; temporarily suppresses
    jubilant's per-call INFO log (set logger to WARNING) to avoid noise.
  • Uses _JubilantModelAdapter for unit/app access.
  • Alias async_wait_for_application_states = wait_for_application_states
    for backward compatibility.

• Rewrite async_get_units() to use get_status_jubilant(); returns a list
  of _JubilantUnitAdapter objects.

• Rewrite async_get_lead_unit() to use get_status_jubilant(); identifies
  the leader via UnitStatus.leader.

• Rewrite async_get_unit_public_address__fallback() to use
  get_status_jubilant() and UnitStatus.public_address.

• Remove tests for the removed async-thread machinery.

• Add tests for the new sync_wrapper behaviour (plain return, coroutine
  drain, nested wrapping).

• Replace all libjuju mock scaffolding with jubilant mocks.

• Tests now patch jubilant.Juju and its .cli() / .status() methods.

• Add _make_jubilant_status() helper to build mock jubilant Status
  objects.

• Rewrite _application_states_setup() to patch get_status_jubilant,
  time.time, and time.sleep instead of libjuju model/connect mocks.

• Update all test_wait_for_application_states_* tests to use the new
  sync polling loop and jubilant mock objects.

• Fix patching targets: resolve_units / block_until_unit_wl_status
  (not their async_ aliases).

• tests/tests.yaml: update magpie-jammy workload-status-message-prefix
  to match current charm output ('Ready, with 1 peer').

• tests/bundles/first.yaml: update top-level series to 'noble'.

Co-authored-by: GitHub Copilot copilot@github.com
Signed-off-by: Felipe Reyes felipe.reyes@canonical.com