Implement statistics::reduce for int96

[little-arhat]

to avoid failing with "not implemented".

[jorgecarleitao]

Thank you for the PR! Could we add a test confirming that the order is correct for timestamps backed by int96?

[jorgecarleitao]

is the order of [u32; 3] the same as its corresponding representation of a timestamp? I.e. don't we need to transform it and then apply the reduce?

[little-arhat]

I'm assuming reading is handled before we get to fold:
deserialize_statistics -> primitive::read::<[u32; 3]> -> types::decode(x) -> T::from_le_bytes.
Then, once we've got NativeType<[u32; 3]> we can use ord it implements ( int96_to_i64_ns(*self).ord(&int96_to_i64_ns(*other))).

Am I missing something?

[little-arhat]

Tests seem to be broken in main?

(r@futu ~/prj/f/parquet2)>>= git br
  feauture-do-not-fail-on-i96-stat
* main
(r@futu ~/prj/f/parquet2)>>= cargo test
   Compiling parquet2 v0.17.0 (/Users/r/prj/f/parquet2)
error[E0063]: missing field `primitive_type` in initializer of `statistics::primitive::PrimitiveStatistics<_>`
   --> src/write/statistics.rs:300:13
    |
300 |             PrimitiveStatistics {
    |             ^^^^^^^^^^^^^^^^^^^ missing `primitive_type`

error[E0063]: missing field `primitive_type` in initializer of `statistics::primitive::PrimitiveStatistics<_>`
   --> src/write/statistics.rs:306:13
    |
306 |             PrimitiveStatistics {
    |             ^^^^^^^^^^^^^^^^^^^ missing `primitive_type`

error[E0063]: missing field `primitive_type` in initializer of `statistics::primitive::PrimitiveStatistics<_>`
   --> src/write/statistics.rs:317:13
    |
317 |             PrimitiveStatistics {
    |             ^^^^^^^^^^^^^^^^^^^ missing `primitive_type`

For more information about this error, try `rustc --explain E0063`.
error: could not compile `parquet2` (lib test) due to 3 previous errors
warning: build failed, waiting for other jobs to finish...
cargo test  5.03s user 0.90s system 392% cpu 1.510 total

Do you have any preferences on how to impl test for int96 timestamps? Should I add completely new series of wite_file/read_ routines, or shall I add new column somewhere?