Add int64 dtype support for RangeFor boundaries

[jim19930609]

Taichi implicitly casts RangeFor boundaries to ti.i32, which is an unnecessary limitation and causes confusion when users explicitly specifies ti.i64 as RangeFor boundaries (#5217).

Ideally, backend's dtype limitation should not bother the frontend design. Unfortunately, we did forbiddened higher precision dtypes in certain cases due to concerns regarding the backends (RangeFor boundaries, ti.atan type conversion, etc..)

We should add a "dtype demotion" pass for each specific backend, and remove all these dtype conversions from the frontend.

[jim19930609]

PR #5322 just added a warning message for this implicit conversion, which did not actually resolve this issue. Ideally we should convert the boundaries to int64 instead of int32.

[lstrgar]

Are there any updates on this? It appears to still be an issue.

[lstrgar]

Bumping this one again! I would love to see this addressed. I would be glad to do it if one of the project maintainers could provide any guidance!

[yojeep]

It's hard to imagine a library that claims to support scientific computing yet doesn't even support Int64.