GH-50057: [C++] Avoid signed overflow in Decimal FromString exponent

[metsw24-max]

Rationale for this change

Decimal{32,64,128,256}::FromString parse the exponent as int32_t, so a string like 0E-2147483648 reaches DecimalFromString/SimpleDecimalFromString with dec.exponent == INT32_MIN and negating it is UB:

decimal.cc: runtime error: negation of -2147483648 cannot be represented in type 'int32_t' (aka 'int')

A near-INT32_MIN exponent overflows the -exponent + fractional_digits addition the same way. Both helpers are reachable from the CSV/JSON readers parsing decimal columns.

What changes are included in this PR?

Compute the scale via internal::SubtractWithOverflow and reject the unrepresentable exponents with Status::Invalid, in both helpers.

Are these changes tested?

Yes, 0E-2147483648 and 1.0E-2147483647 added to DecimalFromStringTest.InvalidInput (runs for Decimal32/64/128/256).

Are there any user-facing changes?

No.

• GitHub Issue: [C++] Signed integer overflow in Decimal FromString exponent handling #50057

[github-actions]

⚠️ GitHub issue #50057 has been automatically assigned in GitHub to PR creator.

[github-actions]

⚠️ GitHub issue #50057 has no components, please add labels for components.

[metsw24-max]

Pushed a clang-format fixup for the C++ Format lint failure, rest of the change is unchanged.

[pitrou]

Thanks for this @metsw24-max , let's just wait for CI

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit b43a417.

There was 1 benchmark result indicating a performance regression:

• Commit Run on test-mac-arm at 2026-06-25 11:27:55Z
  • tpch (R) with engine=arrow, format=parquet, language=R, memory_map=False, query_id=TPCH-02, scale_factor=1

The full Conbench report has more details. It also includes information about 15 possible false positives for unstable benchmarks that are known to sometimes produce them.