test: stop AccountTest failing when the second ticks mid-test

[blaipr]

AccountUseCases::create() and ::updatePassword() stamp passDate with time(). Five
expectations in AccountTest built theirs by calling the same factory, a moment before the
production code called it — so whenever the second ticked between the two, the two models differed
by exactly one and the test failed:

-    'passDate' => 1788098773
+    'passDate' => 1788098774

It is a real failure with nothing wrong in the code, it lands in whichever pull request happens to
be open, and it cannot be reproduced on demand — which is what made it read as a mystery rather
than a bug. It failed the 8.5 unit job on #895 while 8.4 passed on the same commit.

The change

anAccountStampedNow() compares everything else exactly and takes only passDate from the actual
model, after checking it is a timestamp from the last few seconds. A factory that stopped stamping
it, or stamped something else, still fails.

Shown both ways

With sleep(1) injected before the repository write, to make the race happen every time:

	old expectations	new matcher
sleep(1) before the write	5 failures	89 tests OK
service mutated to pass a different DTO	5 failures	5 failures

The second row is what keeps the matcher honest: it is tolerant of the timestamp and of nothing
else. Mutating the factory itself proves nothing here and was discarded — the test builds its
expectation with that factory, so a change to it moves both sides equally.