[master < ] Add python & cpp batching

[antoniofilipovic]

[master < Task] PR

• Check, and update documentation if necessary
• Provide the full content or a guide for the final git message

To keep docs changelog up to date, one more thing to do:

• Write a release note here
• Tag someone from docs team in the comments

[antoniofilipovic]

Some decisions made along the way:

• initializer will receive the same parameters as the main func in batched procedure; thus, we do type checks in mgp.py. The same parameters are sometimes required to do proper initialization, i.e. reading from a file, etc...
• in py_module.cpp there is an option not to invalidate the PyGraph object. This option would give users the option to cache vertices, edges, and other objects from storage and batch return them from the procedure. There are two perspectives in this case:
  1. Tech perspective: we would need two different memory allocators since objects which plan to be cached need to be held in memory until the batched function is called again. Currently, everything is allocated with the same memory resource which after one batch is executed is restarted to a clean state (memory deallocated) to reduce memory usage. In code there are lot of checks if same memory resource is used for PyGraph and PyVertex and these would need decoupling.
  2. User story: Batching works best when it is used for streams and returning of large objects to reduce memory usage. Vertices and edges are not large objects, thus seems unnecessary to batch return of vertices. Other point is that vertices in order to be cached, user would need get all vertices from storage, which would cause upsurge of memory which seems like not necessary.
• [DISCUSS] during the execution of batching procedure there is an option to copy the shared pointer to read lock on all modules if we want for batched procedure not to be restarted (restart requires global lock). It seems like this is also more of a bug than a feature in regular procedures, as the user could write MATCH (n) CALL proc.do_smth() and during execution from another session called mg.load(proc) which would result in issue as only what is checked in code is size of results. Return types are checked so that they suit what was registered.

[Josipmrden]

Some comments but mostly good work

[Josipmrden]

Changes fixed. Minor comment, don't need to re-ask for review again

[andrejtonev]

Just a few comments.

[antoniofilipovic]

@vpavicic for release note:
Add an option to batch results from procedures from both Python and CPP procedures

I will open PR shortly for procedures on docs