[red-knot] gather type prevalence statistics

[carljm]

Something Alex and I threw together during our 1:1 this morning. Allows us to collect statistics on the prevalence of various types in a file, most usefully TODO types or other dynamic types.

[carljm]

Hmm, the test failure here is very odd! Test definitely passes locally. Will have to look into this later.

[AlexWaygood]

that is really weird. it passes for me locally too (and it's obviously passing on all but one of our CI jobs).

That said, the statistic given by the one job that's failing is a statistic that makes more sense intuitively to me...

[carljm]

The issue is that in a release build we (intentionally, for perf reasons) omit tracking of distinct messages / file-and-line information for Todo types. So this results in fewer different kinds of Todo types in a release build. Then this interacts with another bug, which is that we are using FxHashMap::extend wrongly to combine statistics from different scopes (this won't sum the totals when the key exists in both maps). The latter bug only shows up if the same type occurs in more than one scope, and that only occurs in our tests when running in release mode so that all Todo types are the same.

[carljm]

Pushed a commit that fixes statistics merging, and fixes the tests.

It remains the case that we can't add any tests specifically testing that the statistics for "different" Todo types are differentiated, or those tests will fail on a release build. (We could switch any such tests off in release build.) We can still build features making use of this differentiation, but they will only work in debug build.