test: use ISO T-separated expected values to catch silent string mismatch

Per review feedback, changing the expected strings to '0001-01-01T00:00:00'
and '9999-12-31T23:59:59' means that if the datetime parsing regressed and
fell back to a str-based comparison, the mismatch against
str(datetime.datetime(1, 1, 1)) == '0001-01-01 00:00:00' would surface
directly through assert_equal - not only via the log_warning check.

The T-separated form still parses via pd.to_datetime -> us fallback to the
same year-1 / year-9999 instances as before, so the test remains a valid
end-to-end assertion of the fix in sqlmesh/core/test/definition.py.

Signed-off-by: mokashang <shangmengjiajiajia@gmail.com>

Author: mokashang

tests/core/test_test.py

@@ -2964,7 +2964,9 @@ def test_out_of_bounds_nanosecond_timestamp_comparison(mocker: MockerFixture) ->
             )
         }
     )
-    expected = pd.DataFrame({"ts_col": ["0001-01-01 00:00:00", "9999-12-31 23:59:59"]})
+    # Use T separator so a string-only comparison (broken path) would mismatch
+    # against str(datetime.datetime(1, 1, 1)) == "0001-01-01 00:00:00".
+    expected = pd.DataFrame({"ts_col": ["0001-01-01T00:00:00", "9999-12-31T23:59:59"]})
     log_warning = mocker.spy(get_console(), "log_warning")
     test.assert_equal(expected=expected, actual=actual, sort=False)
     for call_args in log_warning.call_args_list: