Draft: Use take-in kernel in repartitioning

[ctsk]

Combined with apache/arrow-rs#7325, tries to use the take_in kernel in repartitioning. The goal is to elide the coalesce step after repartitioning.

[alamb]

Hi @ctsk -- is this PR ready for running some benchmarks?

[ctsk]

@alamb This PR should be able to run benchmarks now. I've added overrides to use the modified version of arrow in the PR and a lockfile to avoid chrono issues. At least it can run tpch :)

[alamb]

I am firing up the benchmarks

[alamb]

I tried to run the clickbench queries using bench.sh and I got an error like this:

Q1: SELECT COUNT(DISTINCT "HitColor"), COUNT(DISTINCT "BrowserCountry"), COUNT(DISTINCT "BrowserLanguage")  FROM hits;
Query 1 iteration 0 took 760.6 ms and returned 1 rows
Query 1 iteration 1 took 785.4 ms and returned 1 rows
Query 1 iteration 2 took 787.9 ms and returned 1 rows
Query 1 iteration 3 took 775.5 ms and returned 1 rows
Query 1 iteration 4 took 786.6 ms and returned 1 rows
Q2: SELECT "BrowserCountry",  COUNT(DISTINCT "SocialNetwork"), COUNT(DISTINCT "HitColor"), COUNT(DISTINCT "BrowserLanguage"), COUNT(DISTINCT "Soci\
alAction") FROM hits GROUP BY 1 ORDER BY 2 DESC LIMIT 10;

thread 'tokio-runtime-worker' panicked at /home/alamb/.cargo/git/checkouts/arrow-rs-583cca34693b79b8/368c1e6/arrow-array/src/builder/mod.rs:509:35\
:
not yet implemented
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Error: Context("Join Error", External(JoinError::Panic(Id(2790), "not yet implemented", ...)))