Dictionary compression

[benjaminwinger]

Note that this depends on #2304

This is implemented inside of StringColumn and StringColumnChunk. With the string column layout being hard-coded into the class, it doesn't really seem to fit the abstractions used for the compression of single chunks.

Implemented in two parts:

1. When copying to the StringColumnChunk, values are de-duplicated as they are stored. This significant memory overhead for data without much duplication, since every string inserted is stored both in the chunk and in an unordered_map that's used to look up the indices of elements already inserted (a copy must be made since references won't survive when the data chunk is resized).
2. Before flushing to the disk, the prepareForFlush method goes over the data again and de-duplicates it. In this case references to the original data are stable and we can use string_view instead of string. This is necessary if we want to remove unused entries, e.g. when strings are read from the disk, existing values are modified, and the original values are no longer used.
3. I've also optimized the scans to avoid both scanning strings multiple times, and copying strings multiple times into a single ValueVector. This is independent of the other parts but speeds up scans for duplicated data.

If the extra memory overhead when copying is a concern, we could remove (1), since (2) is sufficient for on-disk de-duplication, but (1) should speed up copying when there is a lot of duplication (I don't have any benchmarks for this though, and will do some benchmarking and profiling to compare the costs if I have time).

It would also be possible to reduce the memory overhead of (1) by using string_views and updating them to point to the new buffer after a resize, but that will also add some new costs.

[codecov]

Codecov Report

Attention: 4 lines in your changes are missing coverage. Please review.

Comparison is base (7ef3d5c) 90.86% compared to head (92bd3e4) 91.39%.
Report is 13 commits behind head on master.

Files	Patch %	Lines
src/storage/store/string_column_chunk.cpp	92.30%	3 Missing ⚠️
src/common/types/ku_string.cpp	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2408      +/-   ##
==========================================
+ Coverage   90.86%   91.39%   +0.52%     
==========================================
  Files        1023     1023              
  Lines       37044    37284     +240     
==========================================
+ Hits        33659    34074     +415     
+ Misses       3385     3210     -175     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.