Fix timestamp parsing to always include milliseconds

[hanouticelina]

fixes #2671.

See the original slack thread (internal) for more context.

The client expects timestamps to follow the pattern %Y-%m-%dT%H:%M:%S.%fZ, but the server was inconsistently returning timestamps without milliseconds when they were zero (e.g. "2024-11-20T20:01:58Z"). This causes the parsing error reported in the issue mentioned above.
we're handling this client-side as it's intentional behavior in the API to reduce precision, see this comment (private).

this is fixed by a small refactoring of parse_datetime() function and adding a normalization to ensure it has six digits of microseconds by padding with zeros or truncating. This approach handles inputs with or without milliseconds and nanoseconds.

cc @ErikKaum

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Wauplin]

Thanks @hanouticelina! Looks good to me. Glad we've already had some unit tests for the existing part ^^

[ErikKaum]

Thank you for handling this @hanouticelina 🙌