Feat: Parallel processing of Dicom to Nifti conversion instead of serial conversion.

[LalithShiyam]

Issue:
The current DICOM-to-NIfTI conversion process is serialized, making it highly time-consuming for large datasets. This bottleneck significantly hampers inference workflows on large-scale data.

Proposed Solution:
Implement parallel processing for DICOM-to-NIfTI conversion to fully utilize all available CPU cores. The recommended options are:

• Dask: Leverage its parallel computing capabilities.
• Multiprocessing: Use Python's multiprocessing library for fine-grained control.

Challenges with Dask Multiprocessing:
Attempting to use Dask's processes scheduler resulted in the following error:

TypeError: cannot pickle '_thread.RLock' object

This error occurs because Dask's multiprocessing scheduler requires objects to be serializable, and some components (e.g., output_manager) cannot be pickled due to underlying locks (_thread.RLock). We can skip the output_manager, but it is needed for the beautiful console updates. And dask's progress bar is not that great to look at.

Next Steps:

• Either stick with Dask's default threads scheduler for I/O-bound tasks, avoiding serialization issues.
• Or carefully refactor the code to use Python's multiprocessing for CPU-bound workloads, ensuring all shared objects are serializable.

[LalithShiyam]

@Keyn34 I will not have the bandwidth to work on this. As I have other fires to put off. Kindly look into this, whenever you have time.

[Keyn34]

We can also close this, right (#169)?