Writing empty or all-null parquet column chunks creates an empty dictionary page

**Describe the bug**

Not sure if this is really a bug, but it is a bit unexpected and has lead to compatibility issues before (, now fixed). It also adds a little bit of metadata overhead, but that is probably not significant.

**To Reproduce**

Can be reproduced as a testcase in `parquet/src/column/writer/mod.rs`:
```rust
#[test]
fn test_column_writer_default_encoding_support_int32_empty() {
check_encoding_write_support::(
WriterVersion::PARQUET_1_0,
true,
&[],
Some(0), // dictionary page offset is set
&[Encoding::PLAIN, Encoding::RLE],
);
}
```

**Expected behavior**

To be discussed. There is actually a benefit to the current behavior when looking at `PageEncodingStats` (#7354), if one column chunk contains only nulls, and all other column chunks are dictionary encoded then the whole column could still be considered dictionary encoded.

**Additional context**

Start with the test case in parquet/src/column/writer/mod.rs and the check_encoding_write_support call for an empty Int32 column. Trace how the dictionary page offset and Encoding values are produced, then confirm the intended behavior for empty or all-null chunks, including the PageEncodingStats consideration. Done means the expected behavior is agreed and covered by a passing regression test.

Written by the indexing model from the issue text.