fix(python): round sub-millisecond durations in to_timedelta_int (#1668)

* fix(python): round sub-millisecond durations in to_timedelta_int

A timedelta carries microsecond precision, so `x.total_seconds() * 1000.0`
in `to_timedelta_int` can be a non-integer float (e.g. 1.234). The old
`is_integer()` assert then failed on such durations and aborted SessionEvent
serialization. Round to the nearest whole millisecond instead.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* docs(python): clarify banker's rounding in to_timedelta_int comment

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: idryzhov

python/copilot/generated/session_events.py

@@ -2538,8 +2538,9 @@ export function generatePythonSessionEventsCode(schema: JSONSchema7): string {
             out.push(`def to_timedelta_int(x: timedelta) -> int:`);
             out.push(`    assert isinstance(x, timedelta)`);
             out.push(`    milliseconds = x.total_seconds() * 1000.0`);
-            out.push(`    assert milliseconds.is_integer()`);
-            out.push(`    return int(milliseconds)`);
+            out.push(`    # Durations can carry sub-millisecond precision; round to the nearest whole ms`);
+            out.push(`    # using Python's default banker's rounding (round-half-to-even).`);
+            out.push(`    return round(milliseconds)`);
             out.push(``);
             out.push(``);
         }