fix(tests): anchor the energydataservice fixture to Predbat's day, not the host's (#5013)

test_energydataservice built its rate fixture from datetime.now() - the machine's clock -
while reading the timezone offset two lines later from my_predbat.midnight_utc, which is
local midnight in Predbat's configured timezone. Between 23:00 and 00:00 UTC under BST
those land on different dates, so the published rates are a day out from the day the code
is modelling and rates[1440] does not exist: KeyError: 1440.

This is not intermittent. The nightly journal flush runs at local midnight, which during
BST is 23:0x UTC, so it lands in the failing hour every night - PR #5011 is a docs-only
change that failed CI for exactly this reason, at 23:12 UTC.

Same trap and same fix as PR #4998 for the multi-car IOG tests, and the trap already has
a bullet in the debug journal ("Build test dates from Predbat's clock, not the machine's").

Verified by running the suite with TZ=Pacific/Pitcairn, which puts the host a calendar day
behind Europe/London: energydataservice reproduces the KeyError before the change and
passes after, as does the full quick suite (82.98s, 4 slow tests skipped). It also still
passes on an unshifted clock.

The suite aborts on the first uncaught exception, so this was masking anything later in
the run; with it fixed, nothing else in the quick suite fails in that window. Nineteen
other test files call datetime.now(), none of them currently affected, but the pattern is
latent - a scheduled CI run in a shifted timezone would catch the next one before a
maintainer does.

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

Author: 3 people

apps/predbat/tests/test_energydataservice.py

@@ -19,8 +19,14 @@ def test_energydataservice(my_predbat):
     failed = 0
 
     print("Test energy data service")
-    today = datetime.now().strftime("%Y-%m-%d")
-    tomorrow = (datetime.now() + timedelta(days=1)).strftime("%Y-%m-%d")
+    # Anchor the fixture to Predbat's own day, not the host's. midnight_utc is local midnight
+    # in the configured timezone (Europe/London in the harness); datetime.now() is the machine's
+    # clock. Between 23:00 and 00:00 UTC under BST those land on different dates, the published
+    # rates are then a day out from the day the code is modelling, and rates[1440] does not
+    # exist - failing this test for the one hour a day CI happens to run in that window. Same
+    # trap, same fix as PR #4998 for the multi-car IOG tests.
+    today = my_predbat.midnight_utc.strftime("%Y-%m-%d")
+    tomorrow = (my_predbat.midnight_utc + timedelta(days=1)).strftime("%Y-%m-%d")
     tz_offset = int(my_predbat.midnight_utc.tzinfo.utcoffset(my_predbat.midnight_utc).total_seconds() / 3600)
     tz_offset = f"{tz_offset:02d}"