fix pr time coordinate handling

[SBeairsto]

Summary

This PR fixes issues with converting forecast data that has separate forecast_initial_time and forecast_hour dimensions into a flattened time coordinate, and resolves a related bug in the zarr-to-torch conversion pipeline.

Changes

1. Fixed time flattening in metadata.py

Problem: When converting forecast data with forecast_initial_time and forecast_hour dimensions to valid time coordinates, the code was:

• Converting hours to nanoseconds instead of hours (timedelta64[ns] → timedelta64[h])
• Creating a MultiIndex that couldn't be properly overwritten
• Had a duplicate line for coordinate assignment

Solution:

init, hour = xr.broadcast(ds.forecast_initial_time, ds.forecast_hour)
valid_time = init + hour.astype("timedelta64[h]")  # Changed from [ns] to [h]
ds = ds.stack(time=("forecast_initial_time", "forecast_hour"))
ds = ds.reset_index("time", drop=True)  # Reset MultiIndex before assignment
ds["time"] = valid_time.stack(time=("forecast_initial_time", "forecast_hour")).values
return ds

This properly converts forecast hours (e.g., 1-12) to hour offsets and creates valid datetime coordinates (e.g., 1979-01-01T07:00:00 instead of 1979-01-01T06:00:00.000000001).

2. Fixed iterator validation in zarr_to_torch.py

Problem: progress_starmap was receiving a zip iterator which couldn't be validated even with the total parameter specified, causing:

ValueError: If the tasks are an iterator, the total parameter must be specified` 

Solution:

pool_tuple = list(zip(  # Convert iterator to list
    indices,
    partial_paths,
    ds[var.name].transpose(*[dim for dim in chunks.keys() if dim in ds[var.name].dims]),
))

progress_starmap(
    parallel_loop, pool_tuple, total=len(pool_tuple), n_cpu=24, chunk_size=1
)

Testing

• Verified time coordinates are correctly computed for forecast data with hour offsets
• Confirmed zarr-to-torch conversion completes without errors
• Validated temporal alignment between datasets using correlation analysis (no offset detected)

Files Modified

• metadata.py: Fixed time coordinate flattening logic
• tools/zarr_to_torch.py: Converted zip iterator to list for progress_starmap