fix validate_and_prepare_single_dict_task by separating "past only" and "future known" cov keys

[HarvestStars]

Closes #345

Hi @abdulfatir
Here is the bugfix about the function "validate_and_prepare_single_dict_task", which had 2 issue points:

1. Originally, one of this func return, the "task_n_future_covariates", will return the ["past only" + "future known"]covariates number, by task_n_future_covariates = len(task_future_covariates_list) as task_future_covariates_list is filled by for key in task_covariates_keys
2. The code seems not to guarantee the last "future known" rows are atcually what we expected, even there is a sorted option.

So, this PR fixed them by separating "past only" and "future known" covs from the "past_covariates" input, and explicitly put the "past only" covs rows above "future known" cov rows, supported by a temp list "ordered_covariate_keys".

[abdulfatir]

@HarvestStars Thanks a lot for catching and fixing this! Left a few minor comments. Could you please address those? I will also run some evals on my side before merging to ensure that we did not break anything accidentally.

[HarvestStars]

Hi @abdulfatir
Thanks for replying so fast, as well the comment for name standarization.
I think they make much sense, so I fixed them following your comment.
Please double check.

[abdulfatir]

@HarvestStars Thanks that looks good to me. Could you also please format the code? You can do that by running.

uvx ruff format src

See: https://docs.astral.sh/uv/guides/tools/

[abdulfatir]

I re-ran the evals and numbers match. In any case, this issue mainly affects fine-tuning. Thank you for finding this and please feel free to open other issues you encounter with fine-tuning. We have not rigorously tested fine-tuning support yet, that's why it has been marked as experimental.

[HarvestStars]

Hi @abdulfatir
I've already done the format by running
uvx ruff format src
And also update the branch, please check.