Multithreaded implementations using OhMyThreads

[lkdvos]

This is an attempt at providing a unifying interface for various multi-threaded approaches to looping over blocksectors, and looping over fusiontrees.

It can be controlled through the scheduler kwarg in various places, upon which OhMyThreads dispatches, such that there is a possibility of creating and experimenting with scheduler types if one so pleases.
By default, there is default_scheduler(tensortype))=SerialScheduler(), which just mimicks the non-multithreaded approaches at minimal overhead.
Various built-in schedulers exist for OhMyThreads, with Greedy, Dynamic or Static schemes, each with up- and downsides depending on the specific case as well as spacetypes. The hope is that this allows a flexible approach that enables users to tweak these settings if they so please, without having to deal with this when this is not necessary.

To do:

• Documentation
• Proper benchmarks
• Abstraction layer for looping over blocksectors, or looping over fusiontrees
• "Atomic Int" based scheduler + task sorting

[codecov]

Codecov Report

Attention: Patch coverage is 87.17949% with 10 lines in your changes are missing coverage. Please review.

Project coverage is 81.59%. Comparing base (370dd92) to head (2a2292b).

Files	Patch %	Lines
src/tensors/vectorinterface.jl	59.09%	9 Missing ⚠️
src/tensors/indexmanipulations.jl	94.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #117      +/-   ##
==========================================
+ Coverage   81.46%   81.59%   +0.13%     
==========================================
  Files          42       42              
  Lines        5588     5574      -14     
==========================================
- Hits         4552     4548       -4     
+ Misses       1036     1026      -10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Jutho]

Maybe the default_scheduler should also take in the specific function as an argument, so that one can define

default_scheduler(::Type{<:TensorMap}, ::typeof(tsvd!))

In principle, the scheduler keyword argument should also be added to tsvd etc, unless we require people that want to use this, to use the more expert-mode mutating methods.