test: give the debug cases and annual integration their own predbat instance (#4530)

The debug regression cases planned against the shared instance, and read_debug_yaml only restores the
attributes its dump actually carries - so anything the dump omits inherited whatever the previous test
left behind. That made the golden plans depend on test ordering, and made the plan produced inside the
suite differ from the one `--debug <case>` produces standalone. Both have caused real confusion: the
goldens have had to be regenerated simply because the registry order changed, and a case that passes
alone could fail in the suite.

Each case now gets a freshly created instance, built the same way the standalone path builds one. The
three now agree exactly - verified that predbat_debug_pre_saving1 produces an identical plan standalone,
running debug_cases alone, and in the full suite.

Doing that removed a mask. debug_cases had been overwriting most of the shared instance immediately
before annual_integration ran, and annual_integration never sets up the state it plans against; left
with the ambient state instead it went from 34s to 463s - silently, still passing. It gets a fresh
instance for the same reason, which also makes it faster than it was before any of this (55s -> 34s).

Full suite 144.77s -> 119.86s, and it no longer depends on what ran before what.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: springfall2008

apps/predbat/unit_test.py

@@ -220,8 +220,14 @@
 
 
 def run_debug_cases(my_predbat):
-    """
-    Run debug case files from the cases directory
+    """Run debug case files from the cases directory.
+
+    my_predbat is deliberately unused: each case gets a freshly created instance instead. read_debug_yaml
+    only restores the attributes its dump actually carries, so on a shared instance anything the dump omits
+    inherits whatever the previous test happened to leave behind - which made these cases depend on test
+    ordering, and made the plan produced here differ from the one `--debug <case>` produces standalone.
+    Neither is a property a golden regression test can afford. Building the instance the same way the
+    standalone path does makes the two agree and makes the result independent of what ran before.
     """
     failed = False
     print("**** Running debug case files ****")
@@ -235,8 +241,9 @@ def run_debug_cases(my_predbat):
         pathname = os.path.dirname(filename)
         if basename == "random_scenarios.yaml":
             continue  # Skip the random scenarios template file
-        test_failed = run_single_debug(basename, my_predbat, filename, pathname + "/" + basename + ".expected.json")
-        total_calculate_plan_time += getattr(my_predbat, "last_calculate_plan_time", 0.0)
+        case_predbat = create_predbat()
+        test_failed = run_single_debug(basename, case_predbat, filename, pathname + "/" + basename + ".expected.json")
+        total_calculate_plan_time += getattr(case_predbat, "last_calculate_plan_time", 0.0)
         case_count += 1
         if test_failed:
             print(f"**** Debug case {basename}: FAILED ****")
@@ -251,6 +258,19 @@ def run_debug_cases(my_predbat):
     return failed
 
 
+def run_annual_integration_isolated(my_predbat):
+    """Run the annual integration test against a freshly created instance.
+
+    my_predbat is unused, for the same reason run_debug_cases ignores it: this test plans a year of
+    sampled days against whatever state the shared instance is carrying, and never sets that state up
+    itself. It used to be shielded by debug_cases running immediately before it and overwriting most of
+    the instance from a debug dump; once debug_cases stopped mutating the shared instance, the ambient
+    state it inherited instead made it 13x slower (34s -> 463s) without ever failing, which is exactly
+    the kind of coupling a test suite should not have.
+    """
+    return test_annual_integration(create_predbat())
+
+
 def create_predbat():
     my_predbat = PredBat()
     my_predbat.states = {}
@@ -499,7 +519,7 @@ def main():
         ("annual_store", test_annual_store, "Annual run store tests", False),
         ("annual_costs", test_annual_costs, "Annual install cost and payback model tests", False),
         ("tariff_catalogue", test_tariff_catalogue, "Tariff catalogue tests", False),
-        ("annual_integration", test_annual_integration, "Annual prediction integration tests", True),
+        ("annual_integration", run_annual_integration_isolated, "Annual prediction integration tests", True),
         ("load_ml", test_load_ml, "ML Load Forecaster tests (MLP, training, persistence, validation)", True),
     ]