Increase speed and parallelism of the limit algorithm and implement descending sorting

[nils-braun]

This PR introduces four changes:

• it simplifies the logic for getting a piece of the dask dataframe starting at OFFSET and ending at LIMIT.
• this allows to not needing to calculate the dask dataframe before, but really only at the actual calculation. The calculated used to be needed for getting the partition boundaries, but they can now also only be calculated "lazily"
• as there is no need to do a full recalculation, I introduced a quick shortcut: if the first partition is already enough, just return this. This is a very typical use case when you just do a "LIMIT 10".
• it allows for descending sorting also in the first column

[mrocklin]

You may want to become familiar with the df.partitions iterable.

dd.concat(df.partitions[::-1])

Going between delayed and dataframes introduces some inconvenience in graph-handling.

[nils-braun]

Uh nice! That looks awesome. Yes, this is definitely much better.

[mrocklin]

At some point I think that we'll want to go through and remove all of the to/from_delayed calls. This isn't critical, but it's nice to do from a performance standpoint (happy to go into this in more depth if you like).

[nils-braun]

Thanks for your advice @mrocklin! Really appreciated. I would actually be really interested in your thoughts about this part.

Probably you see it right away, but what I would like to achieve is basically two things

1. call a function on every partition, while also having the partition index as an input argument
2. calculate partition_borders only when executing the DAG - not already when creating it (this is what I have done before this PR, and it slows down the process).

So far, the delayed function was all I could come up with, but I definitely think there is a better way.

[nils-braun]

I have created an issue to fix this, so we can go on with this PR.

[mrocklin]

len(df.partitions[0])

[codecov-io]

Codecov Report

Merging #75 (2bc8920) into main (8abc48c) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main       #75   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           34        34           
  Lines         1383      1383           
  Branches       185       189    +4     
=========================================
  Hits          1383      1383           

Impacted Files	Coverage Δ
dask_sql/physical/rel/logical/sort.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8abc48c...2bc8920. Read the comment docs.