Land the two changes that merged into a branch instead of master

[FrayxRulez]

#56 and #58 never reached master. My fault: I opened both against unicode-classes rather than master, so when they were merged they went into that branch, which is where they still are. GitHub reports them as merged, and they are — just not here.

This replays the two commits onto current master, without disturbing what has landed since. It carries no grammars.dat: master's was regenerated after #55 and #57, and the bot will regenerate it again for this, because generate.js is in the workflow's path filter.

• generate.js from Escape characters a pattern cannot carry as bytes #58 — escapeNonAscii in sanitize, and writeString encoding UTF-8 rather than one byte per code unit. Cherry-picked onto master's copy, so Give an alias the language title when it has none of its own #57's alias-title fallback is untouched.
• .github/workflows/check-grammars.yml and tools/ from Check every pattern against both regex engines on a pull request #56 — the escape allowlist, both engines compiling every pattern, and the diff of what each one matches.

Merging unicode-classes directly would not have worked: it predates #57 and carries an older grammars.dat, so it would have reverted both.

Once this is in, the .dat the bot produces should drop the last 33 patterns holding a truncated byte, and the check this adds will be running on every pull request that touches a grammar.

Please delete unicode-classes, check-grammars and write-string afterwards, so the merged-but-stranded commits cannot be mistaken for live branches.

🤖 Generated with Claude Code