Constant compression

[benjaminwinger]

Constant compressed columns use INVALID_PAGE_IDX as their page index in the metadata, and reading from them uses the normal functions, but there's a special case for when we try to read from an INVALID_PAGE_IDX.

To get it working for nulls I had to set up out of place commits for the null column. This always happens when the parent column does an out of place commit (unnecessary, but it would take a lot of work to de-couple that I think), but we may now need to do an out-of-place commit for the null column when doing an in-place commit on its parent column. As such, I've removed in-place updates to the null column when we do in-place updates to the parent column, and have it otherwise doing an out-of-place update on the null column only if the main column does an in-place update to avoid duplicating work.

I'd added some bounds checking, and more fine-grained control over the compression choices at the ColumChunk level when debugging, which I've left in, but could be removed since I was able to get it working for every fixed-sized type.

[codecov]

Codecov Report

Attention: 26 lines in your changes are missing coverage. Please review.

Comparison is base (ad5a79f) 92.83% compared to head (9147b85) 92.91%.

Files	Patch %	Lines
src/storage/store/column.cpp	89.79%	10 Missing ⚠️
src/storage/compression/compression.cpp	86.53%	7 Missing ⚠️
src/include/storage/compression/compression.h	58.33%	5 Missing ⚠️
src/storage/store/column_chunk.cpp	84.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2516      +/-   ##
==========================================
+ Coverage   92.83%   92.91%   +0.07%     
==========================================
  Files        1025     1025              
  Lines       38324    38401      +77     
==========================================
+ Hits        35579    35679     +100     
+ Misses       2745     2722      -23     

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

[benjaminwinger]

Benchmark actually failed, despite the job status. The storage is not compatible with the old version since I'd changed the compression ID numbers to turn them into a bitfield. Maybe that should be changed anyway, since what's stored in the file doesn't need to be a bitfield, and that limits the total number of values. We could have a separate bitfield for configuration, but since it's not being used outside of the ColumnChunk, maybe it's a better idea to revert that part for now and leave more fine-grained compression settings to a later time.