Performance issues of type inference, and tricks

There's a [discussion](https://mail.openjdk.org/pipermail/compiler-dev/2025-January/029197.html) on https://mail.openjdk.org/mailman/listinfo/compiler-dev under the heading "Exponential compilation time for List.of(Map.entry) expressions"

Someone presented an [example program](https://gist.github.com/dweiss/93caf579a89f3dbbfac5c292048e42ad) which makes javac, intellij and also ecj go into deep thought mode for a long time.

Some take-aways from that discussion:
* the core problem is inherent in JLS, or even in the father of all similar type systems: ML. The problem **is** exponential. Period.
* IntelliJ has an [inspection + quickfix](https://www.jetbrains.com/help/inspectopedia/JavacQuirks.html) that let's users add explicit type arguments where it helps to avoid the problem
* javac already performs some tricks to cut down the number of inference variables, see https://mail.openjdk.org/pipermail/compiler-dev/2025-January/029239.html

Items (2) and (3) are worth considering also in JDT.