[ktz03]

Description of change

LlamaJsonSchemaGrammar.parse rejects "😀" for {type: "string", minLength: 1, maxLength: 1} even though the native grammar accepts it. Conversely, parsing accepts the same single code point with minLength: 2, while the native grammar rejects it. The validator uses JavaScript's UTF-16 string length, which counts this character twice.

Count decoded Unicode code points for string length checks and their error messages. Add public grammar/parse regressions for ASCII, BMP characters, astral characters, combining characters, a ZWJ sequence, decoded JSON escape pairs, and object properties. GBNF generation is unchanged.

Validation on Windows with Node.js 24.21.0:

• Five new regression cases fail on master; 25 existing or control cases pass. All 30 grammar tests pass after the fix against current compiled public modules in an isolated package runtime with the prebuilt Vulkan v0.5.0 backend.
• 148 other offline standalone tests pass.
• npm run format, npm run lint:eslint, and npm run test:typescript pass.
• npm run docs:generateTypedoc and the VitePress documentation build pass.
• A separate 31-case public API check passes after the fix.

The full model-dependent suite, other native suites, and live Hugging Face tests were not run. No model inference was performed.

Pull-Request Checklist

• Code is up-to-date with the master branch
• npm run format to apply eslint formatting
• npm run test passes with this change — the validation above passed; the full model-dependent suite was not run
• This pull request links relevant issues as Fixes #0000 — N/A, no existing issue found for this validation error
• There are new or updated unit tests validating the change
• Documentation has been updated to reflect this change — N/A, no API or usage change
• The new commits and pull request title follow conventions explained in pull request guidelines (PRs that do not follow this convention will not be merged)

[giladgd]

Thanks for the PR!
I've pushed a performance optimization so that validating the length against code points will be a cheaper and faster operation