[DataGrid] Filtering performance: use Set() when applicable

[romgrk]

Last part of #9120

edit: Read the summary below, these results are outdated.

This PR improves performance by using Set where applicable.

	Elapsed (ms)
Without Set	56.1 (+-2.5)
With Set	35.6 (+-1.8)

Test: 100,000 rows of the commodity dataset, filter one string column with the string "am". We observe gains of about 35% for this test.

Note that I was thinking we'd need to update the selectors to keep returning objects instead of Sets, but I haven't found any public selector that exposes them, they're all @ignore Do not document. If you see anything that would break semver guarantees, please point it out.

[mui-bot]

Netlify deploy preview

Netlify deploy preview: https://deploy-preview-10068--material-ui-x.netlify.app/

Updated pages

No updates.

These are the results for the performance tests:

Test case	Unit	Min	Max	Median	Mean	σ
Filter 100k rows	ms	-163	127.5	-6.7	-7.86	119.49
Sort 100k rows	ms	943.4	1,701.5	1,606.3	1,436.96	264.51
Select 100k rows	ms	656.3	951.8	885.4	821.44	118.234
Deselect 100k rows	ms	188.9	360.2	218.7	242.92	60.181

Generated by 🚫 dangerJS against 5d1442c

[romgrk]

Summary of this PR:

• I initally did the switch from {} lookup object to Set, and observed the improvements described above.
• I noticed that our {} lookups work as such: false means hidden, true and undefined means not hidden. In other words, the default behavior was to show the row if it hadn't been added in the lookup object. This is the case for autogenerated rows, that not added to the lookup but we always want to show. To comply with those rules, I changed the meaning of sets so that e.g. visibleRows became hiddenRows, and we can use Set to encode things as before.
• However, making that change lost nearly all of the performance improvements observed in the first step.

At this point, I'm not sure if I'm still interested in merging this due to the limited performance benefits that it brings. One change that I do like however is that it becomes clear what the rules are. It took me some time to figure out that tests were failing because both true and undefined meant the same thing.

cc @mui/xgrid

[cherniavskii]

This feels risky TBH. While gridVisibleRowsLookupSelector is not documented, it's exported and not marked as unstable/experimental.
Should we wait with this PR until v7 alpha?

[cherniavskii]

However, making that change lost nearly all of the performance improvements observed in the first step.

I don't understand what caused the loss of the performance improvements.
First, you tried using Set to keep all the visible rows.
Then, you reversed it to only keep the hidden rows. Did this increase the number of records in the Set drastically?
Or is it something else that impacted performance?

[romgrk]

Also didn't make sense to me, but I haven't had enough time to investigate further.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[romgrk]

Although the performance gains are interesting I don't have the bandwidth to continue exploring this one so I'll close it for now. If we look into it again, I'm pretty sure we can expose a backwards compatible interface even if we change the internal model.