Allow NDarray with element counts exceeding Int32 max

[harry-xi]

Overview

Allow NDarray elements to exceed the Int32 maximum (even if no single dimension exceeds the Int32 maximum).

Problem

When I attempted to allocate an ndarray with an element count exceeding the int32 maximum, the program terminated with an Out Of Memory error.

In certain three-dimensional image processing tasks, we do indeed require large three-dimensional matrices with similar structures.

Proposal

• Allow NDarray elements to exceed the Int32 maximum

Evidence

No response

Scope / Non-goals

No response

Benchmark / Performance

No response

Breaking changes

No response

Related issues

No response

[Nucs]

The issue is that we are using int pretty much everywhere and without proper optimization via the new backend (#573) we might will double the compute time as compared to using int.
Regardless of that issue, being long-index compliant is more important and I'll investigate how numpy handles this problem and how we can utilize it.

Thank you for a good first issue.

[Nucs]

Tracked in #584 — migrating all indexing from int to long (breaking change).

The fix involves changing ~120 files: Shape struct fields, IArraySlice, IL kernel generators, iterators, and all public APIs. C#'s implicit int → long conversion means most existing code will still compile, but code storing results like int[] shape = arr.shape will need to change to long[].