fix(agentchat): persist GraphFlow triggered_activation_groups across save/load_state

[nolanchic]

Closes #7043.

Why are these changes needed?

GraphFlow workflows become unrecoverable when paused mid-traversal in a cyclic graph (e.g. A → B with B looping back to A). After save_state → load_state into a fresh team, the resume terminates immediately with "Digraph execution is complete" and never reaches the exit node.

Root cause

GraphFlowManager.save_state / load_state (_digraph_group_chat.py, lines 512-529) persisted message_thread, current_turn, remaining, enqueued_any, and ready — but not _triggered_activation_groups.

_triggered_activation_groups (a Dict[str, Set[str]]) records which activation group caused a node to be enqueued. _reset_triggered_activation_groups (line 438) uses it during select_speaker (line 466) to restore a re-entry node's _remaining count to its original value, so the node can be re-triggered on the next loop iteration.

On resume, _triggered_activation_groups is {} (only initialized in __init__), so _reset_triggered_activation_groups becomes a no-op for the re-entry target. Its _remaining stays at 0 (the value saved mid-traversal) and gets decremented to -1 on the next loop iteration → the node is never re-enqueued → _ready drains → _apply_termination_condition sees not self._ready and emits "Digraph execution is complete" (lines 488-501).

Failure trace (cyclic A→B, B loops to A)

1. Phase 1: run A → B="loop". update_message_thread decrements _remaining["A"]["A"] to 0, enqueues A, records _triggered_activation_groups["A"]={"A"}.
2. MaxMessageTermination(3) fires before select_speaker runs, so _reset_triggered_activation_groups("A") is never called. In-memory state: _triggered={"A":{"A"}}, _remaining["A"]["A"]=0, _ready=["A"].
3. save_state writes _remaining["A"]["A"]=0 and _ready=["A"] but drops _triggered_activation_groups.
4. load_state into a fresh team: _triggered={}, _remaining["A"]["A"]=0, _ready=["A"].
5. Resume: select_speaker pops A, calls _reset_triggered("A") → no-op (A not in the empty dict) → _remaining["A"]["A"] stays 0 → next loop iteration decrements it to -1 → A never re-enqueued → premature "Digraph execution is complete".

Fix

Persist _triggered_activation_groups in save_state (serializing each Set as a list) and restore it in load_state (rebuilding the sets):

# save_state
"triggered_activation_groups": {
    target: list(groups) for target, groups in self._triggered_activation_groups.items()
},
# load_state
self._triggered_activation_groups = {
    target: set(groups) for target, groups in state["triggered_activation_groups"].items()
}

Fields derived purely from graph structure (_activation, _origin_remaining, _edges, _parents) are already reconstructed by __init__ on the new manager and need no persistence — verified they are never mutated at runtime.

Scope

Only cyclic / looping graphs are affected. Linear graphs are unaffected because the reset mechanism only matters for re-entry nodes. The existing test_graph_flow_stateful_pause_and_resume_with_termination (linear A→B) already passed and continues to pass.

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. (None — internal state-serialization fix.)
• I've added tests (if relevant) corresponding to the changes introduced in this PR. (New test_graph_flow_cyclic_pause_resume_reaches_exit: runs one loop iteration → save_state → load_state into a fresh team with fresh agent instances → asserts the exit node C is reached. Verified the test fails on the unfixed code — sources [A, B] with stop_reason \"Digraph execution is complete\" — and passes with the fix — sources [A, B, A, B, C].)
• I've made sure all auto checks have passed. (Local verification below.)

Local verification

cd python
uv run pytest packages/autogen-agentchat/tests/test_group_chat_graph.py -q   # 77 passed (was 76)
uv run ruff format --check packages/autogen-agentchat/src packages/autogen-agentchat/tests   # already formatted
uv run ruff check packages/autogen-agentchat/src packages/autogen-agentchat/tests            # all checks passed
uv run pyright packages/autogen-agentchat/src packages/autogen-agentchat/tests               # 0 errors
uv run mypy --config-file pyproject.toml packages/autogen-agentchat/src packages/autogen-agentchat/tests  # no issues