chore(CI) Upgrade toolchain to Rust-1.87

[kadai0308]

Which issue does this PR close?

• Closes Update workspace / CI to Rust 1.87 #16061 (comment)

Rationale for this change

Improve build times.

What changes are included in this PR?

Updating toolchain to Rust 1.87 and fixing associated errors and warnings.

The main changes including:

• datafusion/common/src/stats.rs:355:9 and the following related changes

359 -             Present(ColumnStatistics),
359 +             Present(Box<ColumnStatistics>),

• datafusion/expr/src/planner.rs:315:1 and the following related changes

317 -     Planned(Expr),
317 +     Planned(Box<Expr>),

• datafusion/optimizer/src/simplify_expressions/expr_simplifier.rs:523:1 and the following related changes

529 -     SimplifyRuntimeError(DataFusionError, Expr),
529 +     SimplifyRuntimeError(DataFusionError, Box<Expr>),

• datafusion/physical-plan/src/aggregates/mod.rs:349:1 and the following related changes

351 -     GroupedHash(GroupedHashAggregateStream),
351 +     GroupedHash(Box<GroupedHashAggregateStream>),

• datafusion/datasource-avro/src/avro_to_arrow/arrow_array_reader.rs:941:29

941 | fn resolve_u8(v: &Value) -> AvroResult<u8> {
    |                             ^^^^^^^^^^^^^^ the `Err`-variant is at least 256 bytes

Are these changes tested?

Are there any user-facing changes?

[alamb]

Thank you for this PR @kadai0308 -- very helpful

I am not sure about the implications of Boxing all the variants -- I worry it will just add additional overhead for planning and API churn (though I'll run some benchmarks0

Rather than adding new Box would you be willing instead to add in the #[clippy(allow(..)) annotations instead for errors about mismatched enum sizes?

I think this would allow us to unblock the upgrade to 1.87 and we can then consider how to fix the enum variant size mismatch as a follow on

For example, one thing I would like to consider is making Expr smaller, for example, this PR: #14366

[kadai0308]

Thank you for this PR @kadai0308 -- very helpful

I am not sure about the implications of Boxing all the variants -- I worry it will just add additional overhead for planning and API churn (though I'll run some benchmarks0

Rather than adding new Box would you be willing instead to add in the #[clippy(allow(..)) annotations instead for errors about mismatched enum sizes?

I think this would allow us to unblock the upgrade to 1.87 and we can then consider how to fix the enum variant size mismatch as a follow on

For example, one thing I would like to consider is making Expr smaller, for example, this PR: #14366

Thanks for the feedback! Yeah, I agree — adding Box to some enum variants and the associated changes could make the codebase and API more unstable and fragile.
I just update the PR with #[allow(clippy::large_enum_variant)].

Another notable change is related to this warning:

941 | fn resolve_u8(v: &Value) -> AvroResult<u8> {
    |                             ^^^^^^^^^^^^^^ the `Err`-variant is at least 256 bytes

It seems we could return Option instead of AvroResult, because

        Value::Array(items) => Ok(Value::Bytes(
            items
                .iter()
                .map(resolve_u8)
                .collect::<Result<Vec<_>, _>>()
                .ok()?,
        )),

If I understand correctly, .ok()? effectively drops the error from AvroResult, making Option sufficient in this context.

[alamb]

Thank you @kadai0308 and @findepi