Add support for chunking

[tdene]

Offloading / onloading the whole tensor is wasteful.

This allows for offloading / onloading of specific chunks.

[fzyzcjy]

Great job! Before digging deeper, may I know a bit of context e.g. where do you want to use it? I am thinking about, shall we use "let user provide num chunks per physical allocation" or "let user provide chunk size per physical allocation", given that some physical allocations may be huge while some are tiny.

[tdene]

Hey fzyzcjy, thanks a lot for the attention!

Sorry that it's taken me some time to get back to you, and that the original PR description is so sparse. I was pushing for a deadline, needed this functionality for the deadline, and wanted a barebones PR up for tracking purposes.

I 100% agree that chunk size is the better way to go about it. And I will share with you the PRs that I need this for, as well as more details, by end-of-week!

[fzyzcjy]

Take your time and looking forward to it!

[fzyzcjy]

@tdene gentle ping :)