fix: Spark cast benchmark shutdown crash

[rui-mo]

The timestamp cast benchmarks captured owning Velox objects in
folly::addBenchmark. Since Folly stores benchmark registrations globally,
these objects could be destroyed after the local pool/context, causing a
shutdown crash in AlignedBuffer::freeToPool().

This PR keeps ownership in main() and makes the registered lambda capture
only non-owning pointers, so Velox objects are destroyed while their
pool/context are still valid.

Stack trace:

EXC_BAD_ACCESS (code=1, address=0x80)

#0  facebook::velox::AlignedBuffer::freeToPool() + 40
#1  facebook::velox::FlatVector<facebook::velox::Timestamp>::~FlatVector() + 288
#2  facebook::velox::RowVector::~RowVector() + 164
#3  std::__1::__function::__func<... addTypedCastBenchmark(...)::$_0 ...>::~__func() + 140
#4  std::__1::function<folly::detail::TimeIterData(unsigned int)>::~function()
#8  folly::detail::BenchmarkRegistration::~BenchmarkRegistration()
#12 std::__1::vector<folly::detail::BenchmarkRegistration>::clear()
#17 folly::detail::BenchmarkingStateBase::~BenchmarkingStateBase()
#19 folly::detail::BenchmarkingState<std::chrono::steady_clock>::~BenchmarkingState()
#20 __cxa_finalize_ranges
#21 exit
#24 start

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	cb4576f
🔍 Latest deploy log	https://app.netlify.com/projects/meta-velox/deploys/6a8880d16cb3ea00080f85e9

[github-actions]

Selective Build Plan

Linux release with adapters is running a selective build of 1 cmake targets (out of 622 total). See the CI workflows README for what this means.

Affected targets (1)

Directly changed (1)

Target	Changed Files
velox_sparksql_benchmarks_cast	CastBenchmark.cpp

---

Fast path • Graph from main@c2a52d02141d