Improve parquet ArrayReaderBuilder tests

As an API suggestion (maybe for another PR, not required):

I would personally find this much easier to understand if the fields were named so I didn't have to refer to `make_int32_page_reader` to figure out what`&[0, 1, 2, 3, 1]` and `&[0, 1, 1, 1, 1]` meant.

Perhaps we could do this with an builder in the test. Something like this perhaps:

```rust
let array_reader_1 = TestArrayReaderBuilder::new()
.with_non_null_i32_values(&[4]) // <-- this is especially non obvous if you are used to normal Arrow arrays
.with_def_levels(&[0, 1, 2, 3, 1])
.with_rep_levels(&[0, 1, 1, 1, 1])
.with_max_def_level(3)
.with_max_rep_level(1)
.build()
```

_Originally posted by @alamb in https://github.com/apache/arrow-rs/pull/9847#discussion_r3173223954_

Start with the ArrayReaderBuilder tests and the make_int32_page_reader helper referenced in the issue and PR discussion. Extract the test setup into a builder whose fields make the non-null values, definition levels, repetition levels, and maximum levels explicit; done means the tests are easier to understand without decoding the numeric arrays from the helper.

Written by the indexing model from the issue text.