Fix merge_kernels benchmark panic due to not wrapping with Scalar

[Jefffrey]

as identified by

• ci: integrate CodSpeed continuous benchmarking #9975

this benchmark was panicking when run, because it was passing a 1-len array as an ArrayRef instead of a Scalar and thus indexing the 2nd/3rd element was causing out of bounds panic; when wrapped in Scalar the single element will be repeated so it wouldn't index out of bounds it would just get the first element

[alamb]

Thanks @Jefffrey -- looks good to me

[alamb]

I verified this panic on main with

cargo bench --bench merge_kernels --profile=dev -- non_null_scalar_vs_array
...
Benchmarking merge_8192_from_i32/non_null_scalar_vs_array/all_true: Warming up for 3.0000 s
thread 'main' (115763431) panicked at arrow-data/src/transform/primitive.rs:31:43:
range end index 8192 out of range for slice of length 1
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

error: bench failed, to rerun pass `-p arrow --bench merge_kernels`

[alamb]

I verified this panic on main with

cargo bench --bench merge_kernels --profile=dev -- array_vs_non_null_scalar
...
Benchmarking merge_8192_from_i32/array_vs_non_null_scalar/99pct_true: Warming up for 3.0000 s
thread 'main' (115764966) panicked at arrow-data/src/transform/primitive.rs:31:43:
range end index 2 out of range for slice of length 1
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

error: bench failed, to rerun pass `-p arrow --bench merge_kernels`