fix generate_series table function overflows

[xiedeyantu]

Which issue does this PR close?

Closes #22208.

Rationale for this change

The table-valued generate_series(start, stop, step) function uses an integer iterator that called current += step without overflow checking. When the series approaches i64::MAX and the step would advance past it, Rust's debug build panics and release build silently wraps, producing incorrect results or a crash visible to end users.

What changes are included in this PR?

• In generate_series.rs, replaced the unchecked *self += step in the SeriesValue for i64 advance method with checked_add, returning a DataFusionError::Execution on overflow instead of panicking.
• Added a regression test in table_functions.slt that verifies SELECT * FROM generate_series(9223372036854775806, 9223372036854775807, 2) returns an execution error rather than panicking.

Are these changes tested?

Yes. A statement error case is added to table_functions.slt covering the exact query from the bug report. The timestamp-based advance implementation already used Option-returning arithmetic and was not affected.

Are there any user-facing changes?

Previously this query would cause a panic (or silent wrap in release mode); it now returns a clean DataFusion error: Execution error: generate_series: integer overflow while advancing series.

[Dandandan]

Interestingly, this is what PostgreSQL returns:

Output:

119 ms
   generate_series   
---------------------
 9223372036854775806
(1 row)

And DuckDB does the same.

┌─────────────────────┐
│   generate_series   │
│        int64        │
├─────────────────────┤
│ 9223372036854775806 │
│ (9.22 quintillion)  │
└─────────────────────┘

I think it makes sense to implement the same behavior (i.e. find max value during instantiation).

[xiedeyantu]

Agreed! I'll align DataFusion's behavior with these database behaviors.

[Dandandan]

Can we make it compatible with PostgreSQL / DuckDB?

[xiedeyantu]

Can we make it compatible with PostgreSQL / DuckDB?

Yes, please wait a moment.

[xiedeyantu]

@Dandandan Please review again. Thanks!

[Dandandan]

I think instead of adding finished we can just update end and also remove the checked_add (use += again since we don't overflow)

[xiedeyantu]

I modified the implementation.