Switch to use u64 range instead of usize, for better wasm32 support

[XiangpengHao]

Which issue does this PR close?

• Closes Support 32-bit Architectures / Replace usize with u64 #5351

Rationale for this change

Ranges should be u64 so that 32 bit platforms can read files larger than 4GB. More details can be found in #5351

What changes are included in this PR?

This is a rather simple change. No functionality change for 64 bit platforms. But is a breaking change for trait implementors.
Given that we already break one in #6619, it's seems like a good timing to also include this change.

This PR added and removed a few casting. I have checked the casting is ok, but please help me check again.

My principle of using usize vs u64:

• Anything already in memory should use usize, e.g., slicing a memory range
• O.w. use u64

Are there any user-facing changes?

[alamb]

The CI failure https://github.com/apache/arrow-rs/actions/runs/12713696201/job/35442156669?pr=6961 should be fixed with this PR:

• Fix WASM CI for Rust 1.84 release #6963

[alamb]

Thanks @XiangpengHao ❤️

Can you help me understand what currently doesn't work with 32 bit builds now?

We have an existing test for WASM32 that seems to work fine: https://github.com/apache/arrow-rs/actions/runs/12680096825/job/35341184364

I also locally tried checking with an i686 target and that worked fine too 🤔

rustup target add i686-unknown-linux-gnu
cd object_store
cargo check --target=i686-unknown-linux-gnu
...

🤔

[alamb]

🤔 maybe it is related to wanting to use the http features?

But it still seems to compile just fine for me 🤔

sudo apt-get install gcc-multilib
rustup target add i686-unknown-linux-gnu
cd object_store
cargo check --target=i686-unknown-linux-gnu --features=http
...

[alamb]

Can you help me understand what currently doesn't work with 32 bit builds now?

For anyone following along, the answer is here: #5351 (comment)

I still think we should try and add some sort of test that would fail on wasm32 before this change and not after the change. If we don't do that I feel like:

1. We are less sure we fixed all the places that require change
2. We may inadvertently introduce regressions but no test would fail

I'll see if I can find some time over the next day or two to help, if no on ebeats me to it