Replace pythonjob task by normal task, drop usage of pickle_node

[superstar54]

In my workstation, it took ~7 mins to run the QE workflow using AiiDA-WorkGraph, while other engines only take ~ 1 mins. The main reason is that the PythonJob task was used to auto-serialize inputs/outputs and preserve data provenance. However, PythonJob is designed to run the Python function on a remote computer. It uploads inputs to a remote folder and retrieves outputs to the local computer. In this project, it's overkill. In the latest aiida-workgraph, the normal keep provenance too. So this PR drops the usage of the PythonJob task.

Another reason is that, the pickled_node task is created for each data node, in this PR, when loading the workflow from a json file, the normal task can also serialize the inputs/outputs in an automatic way, so the pickle_node is not needed anymore. When writing the workflow using AiiDA, we prepare the AiiDA data nodes as inputs, so pickle_node is not needed either.

This PR reduces execution time from ~7 mins to ~ 1.5 mins.

This PR needs aiida-workgraph>=0.5.1.

[GeigerJ2]

Thanks for the nice changes, @superstar54! (both here, as well as in WG itself)
CI runs through now. Had to update the simple workflow to use the new approach, and currently, I'm still installing aiida-workgraph v0.5.1 via pip in the environment.yml file. We can wait with the merge until it's available on conda (thanks, @jan-janssen!), or merge it now already, I don't really have a strong opinion.

[GeigerJ2]

LGTM, and to the 🌔 🚀