Compiler tests diff runner

[gabritto]

No description provided.

[jakebailey]

Comparing CI test time of the runner package before/after AST caching:

• Regular run: 6m19.137s -> 52.906s
• With concurrent programs: 6m30.342s -> 1m8.742s
• In race mode: 2670.664s -> 391.295s

Roughly 6-7x faster.

I still want to get our CI running on better builders for race mode, though. Race mode has about a 5x cost, it seems (race mode does concurrent programs too, of course, so the last two numbers can be directly compared).

[jakebailey]

Switching to a new pool greatly improved the performance. I mistakenly used an 8 core machine instead of a 16 core, and yet it ran race mode in 86 seconds, which is already like 4x faster even though it's only twice the core count of the default GHA runner. I'll rerun with the 16 core I intended later.

[jakebailey]

16 core machine nets 76.860s, not much of an improvement, funnily.

[jakebailey]

Unfortunately after #373, this PR is now crashing on a few tests.

[jakebailey]

There's one failing test: https://github.com/microsoft/TypeScript/blob/main/tests/cases/conformance/externalModules/typeOnly/chained2.ts

Seemingly in concurrent programs, not all checkers have a symbol at the location for A or B in d.ts.

[gabritto]

Submodule test mappedTypeRecursiveInference.ts relies on the no truncation truncation limit of the type printer (noTruncationMaximumTruncationLength in Strada), otherwise it attempts to print forever. Since we don't have the actual type printer yet, I'm disabling it for now since that's easiest.