Explicit variance annotations for built-in .d.ts files

While profiling some projects using typeslayer, I've noticed that according to the type checking traces some time is spent calculating the type variance of common types like `Promise` and others.

I've explicitly annotated some of these generic types, and I've verified that it that sped up typechecking for some of my complex use-cases.
I've verified that these speed-ups also translate to `tsgo` by checking with hyperfine.

I'd like to ask if a PR to add explicit type variance to some of the most commonly used library types could be a valuable addition? I could set up a PR so some performance tests could be run to see if it works in aggregate, I just wanted to ask if it's something that aligns with the projects' goals and to maybe understand if this was tried already

Based on my limited understanding the pros would be:
- More people learning about variance and explicit variance annotations
- Slightly increased performance in some complex scenarios

The cons would be:
- Different error messages as now we'd see `super` and `sub` prefixes that might confuse users
- larger size on the wire/disk
- higher overhead to parse

No source files or tests are named. Start by reviewing the built-in .d.ts library types and the type-checking traces described in the issue, then compare tsgo measurements with hyperfine. Done means agreeing on the target commonly used types and demonstrating the aggregate performance and diagnostic impact with performance tests.

Written by the indexing model from the issue text.