[ty] Fix panic when attempting to provide autocompletions for an instance of a class that assigns attributes to self[0]

[AlexWaygood]

Summary

Ty currently panics when attempting to provide autocompletions in this scenario:

class Point:
    def orthogonal_direction(self):
        self[0].is_zero


def test_point(p2: Point):
    p2.<CURSOR>

The root cause of this issue is that we're incorrectly registering the [0].is_zero place as an "instance attribute" Place. Places should only be registered as instance attributes if they have the form self.<NAME>; self<SUBSCRIPT>.<NAME> does not constitute an instance attribute.

This panic was uncovered by #18705.

Test Plan

I added the above repro as a test to ty_ide/completions.rs

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[sharkdp]

Thank you!

I wondered if matches!(place_expr.sub_segments(), &[PlaceExprSubSegment::Member(_)]) should be a method on PlaceExpr, but it already has PlaceExpr::is_instance_attribute and PlaceExpr::is_instance_attribute_named with subtly different meanings. But maybe that's an argument for doing it (and clarifying the difference in doc comments)?

[AlexWaygood]

Thank you!

I wondered if matches!(place_expr.sub_segments(), &[PlaceExprSubSegment::Member(_)]) should be a method on PlaceExpr, but it already has PlaceExpr::is_instance_attribute and PlaceExpr::is_instance_attribute_named with subtly different meanings. But maybe that's an argument for doing it (and clarifying the difference in doc comments)?

Good call. I added some higher-level abstractions to place.rs, and lots of doc-comments to distinguish semantic-index-internal methods from crate-public API 👍

[sharkdp]

Thank you for the update, this is great!