fix(python): fix event routing sender mismatch for playhead events

[richardssam]

Linked issues

N/A

Summarize your change.

Fixes EventToPythonThreadLockerActor so that playhead and selection events
subscribed from Python are actually delivered, and prevents a SIGSEGV crash
when the subscribed actor is torn down during a timeline switch.

Describe the reason for the change.

When a C++ event group (e.g. a Playhead's broadcast group) sends a message, the
CAF sender is the owner actor (the Playhead itself), not the event-group
actor that the Python code subscribed to. EventToPythonThreadLockerActor
looked up callbacks by sender address, found no match, and silently dropped the
message. This meant subscribe_to_playhead_events and similar subscriptions
delivered nothing.

A second problem: the broadcast_down_atom handler was a TODO no-op. When the
subscribed owner actor died (e.g. on a timeline switch), any in-flight message
attributed to its address was dispatched to a now-invalid callback_func,
causing a signal 11 (SIGSEGV) crash.

Describe what you have tested and on which operating system.

Tested on macOS: a Python plugin using subscribe_to_playhead_events now
receives position_atom events on every frame change. Timeline switches (which
tear down and recreate the Playhead actor) no longer crash the application.

Add a list of changes, and note any that might need special attention during the review.

• src/python_module/src/py_context.cpp:

  • Add callback_to_event_group_ map (Uuid → caf::actor) to track which
    event-group actor owns each callback, enabling correct leave_broadcast
    when a subscription is cancelled.
  • Add a new 3-argument (events_source, owner_actor, callback_id) message
    handler that registers owner_actor's address as an additional routing key
    for the same callback, so messages sent by the owner are routed correctly.
  • Implement broadcast_down_atom handler: on actor death, walk
    actor_to_callback_uuid_ for the dead address, erase the corresponding
    entries from callback_to_event_group_, then remove the actor entry.
  • Move the leave_broadcast send into the Uuid-cancellation handler so it
    uses callback_to_event_group_ for O(1) lookup instead of a linear scan.
  • Extend py_add_message_callback to accept an optional third argument
    (owner actor), passing it to the new 3-argument handler when present.

  Note for reviewers: the broadcast_down_atom fix addresses a latent crash
  that would have occurred for any Python subscription whose source actor could
  be destroyed at runtime (playheads, selection actors). The original TODO
  comment acknowledged this was unfinished.

• python/src/xstudio/api/module.py:

  • subscribe_to_event_group passes the owner actor as the third argument to
    add_message_callback when one is available.

If possible, provide screenshots.

N/A

This was developed with the assistance of claude code.

[richardssam]

Sorry, ignore this its a duplicate.