GH-50149: [C++][Parquet] Avoid process abort when encoding fuzzer encounters OOM

[pitrou]

Rationale for this change

In the Parquet encoding fuzzer, an OOM error when trying to roundtrip the encoding payload leads to a hard error that is reported as an issue on OSS-Fuzz. This should be converted to a soft error, i.e. a potential log message but not a process abort.

Are these changes tested?

By additional fuzz regression file as well as manually.

Are there any user-facing changes?

No.

• GitHub Issue: [C++][Parquet] OOM leads to hard failure in parquet-encoding-fuzz #50149

[pitrou]

@github-actions crossbow submit fuzz

[github-actions]

Revision: 1cbeef7

Submitted crossbow builds: ursacomputing/crossbow @ actions-19765c5446

Task	Status
test-build-cpp-fuzz

[adamreeve]

👍 makes sense to me

[adamreeve]

Just noticed that there is another behaviour change. Previously any Parquet exceptions would be converted to a status code return value, but now they will get checked below and treated as a fatal error. That seems appropriate though.

[pitrou]

You're right, and that's for the better indeed.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit ed536d1.

There were no benchmark performance regressions. 🎉

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.