Normalize lone surrogates in documentation and symbols

[mindaugasrukas]

Prevent protobuf serialization failures when TypeScript produces lone UTF-16 surrogates in documentation or generated symbol identifiers. This covers string literals such as '\ud800' and prototype property names:

export function C() {}
C.prototype = { '\ud800': 1 }

Use String.prototype.toWellFormed() to replace unmatched surrogates with �, preserving valid Unicode and existing textual escapes. Normalize identifiers centrally in ScipSymbol so definitions, references, and relationships use consistent values. Apply the same treatment to generated signatures, JSDoc, and module documentation.

Add es2024.string to TypeScript's lib configuration because the existing ES2022 declarations do not include toWellFormed(). Without it, TypeScript rejects the new calls as an unknown string method. This adds the required type declarations while keeping the compilation target at ES2022. The project's existing Node.js requirement already provides runtime support, so no polyfill or runtime upgrade is needed.

Regression coverage verifies protobuf serialization and matching property definitions and references. It also checks adjacent surrogates: \ud800\ud800\udc00\udc00 becomes �𐀀�, preserving the valid middle pair.

Validation

• Build passed (yarn build).
• Tests passed (yarn test).
• Formatting checks passed (yarn prettier-check).
• Lint passed (yarn eslint).

[mindaugasrukas]

@eseliger Could you review this PR when you have a chance?

[christoph-sg]

Thank you for the PR! While reviewing I looked at a few more ways for lone surrogates to appear in the generated index. For example this code will produce a Symbol with a lone surrogate in it.

export function C() {}
C.prototype = { '\ud800': 1 };

Let me know if you'd like to extend this PR to cover generated symbols as well, or if you'd like me to take over. In general I'd probably go with String.toWellFormed() instead of trying to escape lone surrogates. They're very unlikely to appear intentionally in source code.

[mindaugasrukas]

Thank you for the PR! While reviewing I looked at a few more ways for lone surrogates to appear in the generated index. For example this code will produce a Symbol with a lone surrogate in it.

export function C() {}
C.prototype = { '\ud800': 1 };

Let me know if you'd like to extend this PR to cover generated symbols as well, or if you'd like me to take over. In general I'd probably go with String.toWellFormed() instead of trying to escape lone surrogates. They're very unlikely to appear intentionally in source code.

Thanks for the review! I’ve extended the fix to cover generated symbol identifiers and switched to toWellFormed() for both symbols and documentation. I also added regression coverage for your prototype example.