STYLE: Replace thread_local variables with m_TensorStoreData data member

[N-Dekker]

It appears preferable that multiple OMEZarrNGFFImageIO instances do not share their TensorStore context and store. With this commit, each OMEZarrNGFFImageIO instance will have its own context and store, stored in its m_TensorStoreData data member.

The m_TensorStoreData data member hides the TensorStore data from the interface, as it is a pointer to a forward-declared struct, similar to the "Pimpl" idiom.

[dzenanz]

Looks good on a glance.

What drives your interest in this remote module?

[N-Dekker]

What drives your interest in this remote module?

Good question 😃 One of our aims for the elastix project is to add OME-Zarr support! At least for the input images. We are hoping to get it working by using your remote module!

[N-Dekker]

So... what inspired you to make this module?

[dzenanz]

Adding support for distributed image registration, as part of a grant-funded project.

[N-Dekker]

Did you write it all from scratch, or is there helpful example code of how to use tensorstore in C++, that you could use?

Like, how did you figure this out?

auto indexDomain = tensorstore::IndexDomainBuilder(dimension).origin(indices).shape(sizes).Finalize().value(); 
  
auto arr = tensorstore::Array(buffer, indexDomain.shape(), tensorstore::c_order); 
auto indexedStore = store | tensorstore::AllDims().SizedInterval(indices, sizes); 
tensorstore::Read(indexedStore, tensorstore::UnownedToShared(arr)).value();

ITKIOOMEZarrNGFF/src/itkOMEZarrNGFFImageIO.cxx

Lines 230 to 234 in 1d0f131

	auto indexDomain = tensorstore::IndexDomainBuilder(dimension).origin(indices).shape(sizes).Finalize().value();
	auto arr = tensorstore::Array(buffer, indexDomain.shape(), tensorstore::c_order);
	auto indexedStore = store | tensorstore::AllDims().SizedInterval(indices, sizes);
	tensorstore::Read(indexedStore, tensorstore::UnownedToShared(arr)).value();

I'm impressed!

[dzenanz]

I started from scratch. I was looking at unit tests in tensorstore, and to a lesser degree at their Python documentation. TensorStore authors also helped me a lot. Take a look at issues opened by me:
https://github.com/google/tensorstore/issues?q=is%3Aissue+author%3Adzenanz+sort%3Acreated-asc
Most of them had pretty long discussions.

[N-Dekker]

I think this pull request is ready 👍 (GitHub does not allow me to approve my own pull request 🤷.)