Prototype type checking w/ pyrefly

[yangdanny97]

This PR prototypes the ability to swap between using ty and Pyrefly as the type checker for Monty, via the MONTY_TYPE_CHECKER env var right now.

The actual diff is small - Pyrefly exposes a minimal API that is easy to call, most of the remaining logic is refactoring to support 2 separate backends. There's just a lot of Cargo.lock churn from pulling in Pyrefly as a dependency.

For context, I believe @samuelcolvin spoke with @migeed-z at PyCon about this, and was open to us experimenting in this direction

It can't be merged in this state, because:

• This points at my own fork of Pyrefly, which prototypes the new API that Monty uses. I will upstream it once we're on the same page.
• Monty depends on a fork of ruff, while Pyrefly depends on a different version of ruff, so right now we're pulling in 2 copies of ruff, and we'll need to decide on a way to reconcile the versions and handle upgrades.
• Needs some profiling and optimization: ATM bench_warm_trivial/bench_warm_builtin are pretty close to ty's performance, but repl-based benchmarks are 3-5x slower (tho admittedly I didn't do much optimization)

I'm putting this PR up just to get some early feedback. Let me know what you guys think!

[codspeed-hq]

Merging this PR will not alter performance

✅ 17 untouched benchmarks
⏩ 15 skipped benchmarks¹

---

_{Comparing yangdanny97:pyrefly (c2c0390) with main (592d757)}

Footnotes

1. 15 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[codecov]

Codecov Report

❌ Patch coverage is 28.84615% with 74 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/monty-type-checking/src/pyrefly_check.rs	0.00%	62 Missing ⚠️
crates/monty-type-checking/src/type_check.rs	71.42%	11 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[yangdanny97]

Ah, the numbers in the original post were outdated, I measured the wrong commit.

The actual speed difference in repl is more like 3x; tho it varies based on the size of the thread pool