fix: write non-ASCII test fixtures as JSON escape sequences

[pikann]

Summary

Fixes the failing CI on the initial commit (run 37125064280): the Test and Golden file matches BlockNote jobs both failed on the line_sep fixture.

Cause

line_sep exercises U+2028 / U+2029 (line and paragraph separators). Those were written as raw characters in testdata/blocknote_fixtures.json, and something in the toolchain stripped them, so the committed fixture silently became plain abc:

• go test converted abc but the golden file expects a
b
c;
• the JS check ran BlockNote on abc and got a different result from the golden file.

Lint passed. Nothing was wrong with the converter itself.

Changes

• Fixtures (testdata/blocknote_fixtures.json): restore line_sep and rewrite the file with every non-ASCII character as a \uXXXX escape, so no editor or tool can strip one again.
• Golden writer (js/golden.test.ts): escape U+2028, U+2029 and U+FEFF when writing testdata/blocknote_golden.json. JSON.stringify emits them raw, which is the same trap. Only two lines of the golden file change (raw characters become escapes); the converted output is identical.
• Guard test (golden_test.go): TestFixturesAreASCII fails if the fixtures contain any raw non-ASCII byte, or if either testdata file contains a raw U+2028, U+2029 or U+FEFF. It immediately caught a second case, a raw U+FEFF in the golden file (the zero-width fixture), which is fixed here too.

Test plan

• go test -race ./...
• golangci-lint run
• cd js && npm run golden (real BlockNote output matches the golden file)
• CI green on this PR (Lint, Test, Golden file matches BlockNote)

🤖 Generated with Claude Code