[ty] Use an interval map for scopes by expression

[MichaReiser]

Summary

The semantic index stores a map from expression to scope because we need to know in which TypeInference (scope) to look up the expression's type. Today, we use a hash map to store the expression-to-scope mapping.

This PR replaces the hash map with an interval map (vector-based) that maps a range of node IDs (expressions) to their corresponding scope. The advantage of an interval map over a hash map is that it reduces memory consumption from O(expressions) to O(~scopes).

The main downside (other than increased complexity) is that the lookup complexity increases from O(1) to O(log(~scopes)). Looking at the benchmark results, the fact that we need to write less data outweighs the slightly slower lookup times.

The instrumented benchmarks show a 1-2% performance improvement. I measured memory consumption on a large project and the overall memory consumption of all semantic indices decreased by about 5%,

Test plan

cargo test

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

Memory usage changes were detected when running on open source projects

flake8 (https://github.com/pycqa/flake8)
-     memo fields = ~66MB
+     memo fields = ~63MB

prefect (https://github.com/PrefectHQ/prefect)
-     memo fields = ~568MB
+     memo fields = ~541MB

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[sharkdp]

Very nice — thank you!

[sharkdp]

Is this something that would change with the proposal in #19271?

[MichaReiser]

Yes, but I don't think it would invalidate the entire approach. Instead, we would have to use a regular sort call in build before building the interval map (and Rust's sorting claims to be pretty good at sorting mostly sorted data)