Fix timezone handling and midnight rounding issues in forecasts

[CodeVishal-17]

Description

This PR fixes issues with timestamp handling in the forecasting pipeline, particularly around
timezone-aware inputs and rounding behavior close to midnight.

Specifically:

• Normalize the input timestamp once in run_forecast, ensuring all downstream logic receives
  a timezone-aware UTC timestamp.
• Handle timezone-aware and timezone-naive inputs consistently (naive timestamps are assumed
  to be UTC with a deprecation warning).
• Fix a rounding edge case in predict_tryolabs where timestamps close to midnight could
  produce inconsistent start_date and start_time values.

This avoids subtle bugs and errors when using timezone-aware datetimes or when forecasts are
run close to day boundaries.

Fixes #320

How Has This Been Tested?

• Manually tested predict_tryolabs with timestamps close to midnight (e.g. 23:40, 23:55)
  to ensure start_date and start_time are derived from the same rounded timestamp.

• Verified that timezone-aware timestamps (non-UTC) are accepted and handled correctly.

• Verified that timezone-naive timestamps continue to work and emit a deprecation warning.

• Checked that behavior for typical daytime timestamps remains unchanged.

• Yes

If your changes affect data processing, have you plotted any changes? i.e. have you done a quick sanity check?

• Yes

Checklist:

• My code follows OCF's coding style guidelines
• I have performed a self-review of my own code
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• I have checked my code and corrected any misspellings

[CodeVishal-17]

Heads up: the failing CI appears unrelated to the timezone changes in this PR.

The errors are coming from API unit tests failing with:
AttributeError: InferenceClient has no attribute chat_completion,
which looks like a breaking change in the HuggingFace InferenceClient API.

Happy to pin the dependency or update the mocks if that’s desired — let me know what you prefer.

[CodeVishal-17]

CI is still failing with:

AttributeError: InferenceClient has no attribute chat_completion

This looks unrelated to the timezone changes in this PR and appears to be due
to a mismatch between the API tests and the current huggingface_hub
InferenceClient API.

I pinned huggingface_hub==0.17.3, but CI still seems to install a version
without chat_completion.

Happy to:
• update the API test mocks, or
• adjust the CI install strategy, or
• leave this to a follow-up PR

— let me know what you’d prefer.