[rust] add node_ext module with full_name methods for constant nodes

[sei40kr]

Summary

• Port Ruby's node_ext.rb functionality to Rust bindings
• Add FullName trait with full_name and full_name_parts methods, implemented for ConstantReadNode, ConstantWriteNode, ConstantTargetNode, ConstantPathNode, and ConstantPathTargetNode
• Add ConstantPathError enum for error handling

Test Coverage

Rust bindings include 11 unit tests compared to Ruby's 8 tests. Additional coverage includes:

• ConstantWriteNode (not tested in Ruby)
• ConstantTargetNode (not tested in Ruby)
• MissingNodes error case (not tested in Ruby)

Public API Changes (cargo-public-api)

Added items to the public API
=============================
+pub enum ruby_prism::ConstantPathError
+pub ruby_prism::ConstantPathError::DynamicParts
+pub ruby_prism::ConstantPathError::MissingNodes
+impl core::clone::Clone for ruby_prism::ConstantPathError
+impl core::cmp::Eq for ruby_prism::ConstantPathError
+impl core::cmp::PartialEq for ruby_prism::ConstantPathError
+impl core::error::Error for ruby_prism::ConstantPathError
+impl core::fmt::Debug for ruby_prism::ConstantPathError
+impl core::fmt::Display for ruby_prism::ConstantPathError
+pub trait ruby_prism::FullName<'pr>
+pub fn ruby_prism::FullName::full_name(&self) -> Result<Vec<u8>, ConstantPathError>
+pub fn ruby_prism::FullName::full_name_parts(&self) -> Result<Vec<&'pr [u8]>, ConstantPathError>
+impl<'pr> ruby_prism::FullName<'pr> for ruby_prism::ConstantReadNode<'pr>
+impl<'pr> ruby_prism::FullName<'pr> for ruby_prism::ConstantWriteNode<'pr>
+impl<'pr> ruby_prism::FullName<'pr> for ruby_prism::ConstantTargetNode<'pr>
+impl<'pr> ruby_prism::FullName<'pr> for ruby_prism::ConstantPathNode<'pr>
+impl<'pr> ruby_prism::FullName<'pr> for ruby_prism::ConstantPathTargetNode<'pr>

Test plan

• bundle exec rake cargo:test passes
• bundle exec rake cargo:lint passes
• Doctests pass

🤖 Generated with Claude Code

[kddnewton]

Why copy on write here?

[sei40kr]

The Cow<'pr, str> was a byproduct of using String::from_utf8_lossy(), which returns Cow<'_, str>. It was not an intentional design choice. Addressed in ec991ae by switching to &'pr [u8], which removed the need for Cow entirely.

[kddnewton]

I don't feel great about utf8 here, it's not guaranteed by the parser (there are ~90 encodings supported).

[sei40kr]

Good point. Fixed in ec991ae by switching from String::from_utf8_lossy / Cow<str> to raw &'pr [u8], so we no longer assume any specific encoding.

[sei40kr]

Thanks for the review!

Why copy on write here?

The Cow<'pr, str> was not an intentional API choice — it was just the return type of String::from_utf8_lossy() leaking through. When the bytes are valid UTF-8 it returns Cow::Borrowed(&str), and when they're not it returns Cow::Owned(String) with replacement characters.

I don't feel great about utf8 here, it's not guaranteed by the parser (there are ~90 encodings supported).

Agreed. I've updated the PR to return raw bytes (&'pr [u8] / Vec<u8>) instead, removing all from_utf8_lossy calls. This is consistent with the existing ConstantId::as_slice() API and avoids silently corrupting non-UTF-8 data.

[kddnewton]

Can we make full_name and full_name_parts into a trait? It is an interface, and it should make the ConstantPathNode implementation simpler.

[sei40kr]

Done in 33b4e6f. Extracted a FullName<'pr> trait with full_name_parts and a default full_name implementation. This simplified the ConstantPathNode impl significantly — replaced the iterative while-loop + reverse() + is_stovetop() pattern with recursive delegation via a full_name_parts_for_node helper.