Add lower bound for groupCubeSize

[Jiaweihu08]

When the number of elements in a table is large, the groupCubeSize used by CubeWeightsBuilder tends to 1. This can leads to a large number of estimated Cube Weight pairs that can cause OOM errors.

For instance, using default values for bufferCapacity and desiredCubeSize, a table with 2600M records split between 1200 partitions, the groupCubeSize is 190. The algorithm estimates 37K cubes, among which only around 1K are used. The workaround is to set a minimum value for groupCubeSize, for instance 1000, so the number of estimated cubes won't be as large. Using this setting, the number of estimated cubes is reduced to about 5K.

Test added:
Using default bufferCapacity and DesiredCubeSize, make sure that the minimum value for groupCubeSize is respected for different numElements.

[Jiaweihu08]

The sampling test failed because now the groupCubeSize is 1000 and there are only a handful of cubes. Use a sampling fraction of 0.01 can solve the problem, should be modify the test? @osopardo1

[osopardo1]

This test in particular only checks if the filtering of the files is done correctly. Yes, is best to do it with a sample of 0.01.

But then we need to test better the number we are using to solve the issue. Or at least, test it in corner cases.

[codecov]

Codecov Report

Merging #107 (888d0f7) into main (55442d9) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main     #107   +/-   ##
=======================================
  Coverage   89.34%   89.35%           
=======================================
  Files          60       60           
  Lines        1286     1287    +1     
  Branches      100      106    +6     
=======================================
+ Hits         1149     1150    +1     
  Misses        137      137           

Impacted Files	Coverage Δ
.../main/scala/io/qbeast/core/model/CubeWeights.scala	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 55442d9...888d0f7. Read the comment docs.

[osopardo1]

Looks good to me!