fix: wrap FieldPath inner Vec in Rc for O(1) cloning

[eikopf]

Summary

• Wraps FieldPath's inner Vec<PathSegment> in Rc so that clone() is O(1) instead of O(n)
• Makes the inner field private, adding a segments() accessor for read access
• Pre-allocates exact capacity in push() instead of clone + push
• Updates two call sites that accessed the inner .0 field directly (eval.rs test and render.rs display impl)

Test plan

• All 566 existing workspace tests pass (cargo test --workspace)
• No public API changes beyond making the inner field private (was only accessed in two internal locations)

Closes #32

🤖 Generated with Claude Code

[eikopf]

It's not clear to me that issue #32 is actually correct? I suspect it's being properly interned by salsa, and if that's not the case then we should address this by interning the path rather than wrapping it with Rc.

[eikopf]

Superseded by #91