fix: improve cross-device evidence and exploration lifecycle

Align Responses cache affinity across execution and summary calls, preserve required checks and evidence uncertainty, and allow safe deletion of inactive explorations while retaining independent reruns.

Validation: backend suite passed except intentional prompt golden drift, then all 17 transcript tests passed after snapshot refresh; 268 frontend tests and production build passed; Ruff, quality ratchet and core type checks passed.

tests/unit/test_cache_affinity.py

@@ -0,0 +1,55 @@
+from concurrent.futures import ThreadPoolExecutor
+from types import SimpleNamespace
+from uuid import uuid4
+import re
+
+from artemis.llm.cache_affinity import cache_affinity_options
+
+
+def context(session, exploration=None):
+    return SimpleNamespace(
+        data_engine=SimpleNamespace(current_session_id=session), exploration_id=exploration
+    )
+
+
+def test_routing_identity_survives_resume_but_isolates_tasks_roles_and_models():
+    ctx = context(uuid4(), "exploration")
+    first = cache_affinity_options(ctx, "flash", "gpt-6-luna")
+    ctx.data_engine.current_session_id = uuid4()
+    assert first == cache_affinity_options(ctx, "flash", "gpt-6-luna")
+    key = first["prompt_cache_key"]
+    assert re.fullmatch(r"[-0-9A-Za-z._:]{1,128}", key)
+    assert "exploration" not in key
+    assert first["extra_headers"] == {"Thread-Id": key, "Session-Id": key}
+    for role, model in [("checker", "gpt-6-luna"), ("flash", "gpt-6.1-sol")]:
+        assert cache_affinity_options(ctx, role, model)["prompt_cache_key"] != key
+    assert (
+        cache_affinity_options(context("another"), "flash", "gpt-6-luna")["prompt_cache_key"] != key
+    )
+
+
+def test_explicit_options_are_preserved_without_mutating_headers_or_adding_aliases():
+    headers = {"thread_ID": "explicit-thread", "session-id": "explicit-session", "X-Test": "ok"}
+    options = {"prompt_cache_key": "explicit.key", "extra_headers": headers, "temperature": 0}
+    actual = cache_affinity_options(context("task"), "flash", "test", options)
+    assert actual == options and actual is not options
+    assert actual["extra_headers"] is not headers
+    assert options["extra_headers"] == headers
+    assert cache_affinity_options(None, "flash", "test") == {}
+    for key in (None, "", "a" * 129, "unsafe\r\nheader"):
+        assert cache_affinity_options(
+            context("task"), "flash", "test", {"prompt_cache_key": key}
+        ) == {"prompt_cache_key": key}
+
+
+def test_concurrent_calls_do_not_share_or_mutate_task_headers():
+    shared = {"extra_headers": {"X-Test": "shared"}}
+    with ThreadPoolExecutor(max_workers=2) as pool:
+        results = list(
+            pool.map(
+                lambda task: cache_affinity_options(context(task), "flash", "test", shared),
+                ["task-a", "task-b"],
+            )
+        )
+    assert results[0]["extra_headers"]["Thread-Id"] != results[1]["extra_headers"]["Thread-Id"]
+    assert shared == {"extra_headers": {"X-Test": "shared"}}