Fix tokenizer handling of escaped identifiers - #2123

Fixes #1349

Reproduction

Before this change, tokenizing the issue examples split one escaped identifier into many word tokens:

const tokenizer = require('./lib/tokenize')
const { Input } = require('./lib/postcss')
let input = String.raw`C\(\#0280ae\)`

Current main produced C, \(, \#, 0280ae, \) as separate words instead of one identifier token.

Root cause

The word scanner stopped at every backslash because RE_WORD_END treats \ as a word boundary. The dedicated backslash branch consumed only the escape sequence itself, so escaped punctuation inside identifiers could never stay attached to the surrounding identifier text.

Fix

Reuse the existing escape-consumption rules while scanning word endings. When the scanner reaches a valid CSS escape, it skips over that escape and continues the same word token; invalid escapes keep the previous token boundaries. This also keeps consecutive hexadecimal escapes in the same identifier token.

Compatibility

No new dependencies. Parsing/stringification behavior is unchanged for normal CSS; the tokenizer now matches CSS identifier semantics for valid escaped punctuation.

Validation

• node -r ts-node/register/transpile-only test/tokenize.test.js: 34/34 passed.
• Regression test with lib/tokenize.js reverted: failed as expected (keeps escaped punctuation in identifiers).
• pnpm run test:coverage: 682/682 unit tests passed.
• pnpm run test:lint: passed with 0 errors and one existing warning in test/visitor.test.ts.
• pnpm run test:types: passed.
• pnpm run test:version: passed.
• pnpm run test:integration: 33 real-world CSS fixtures passed.
• pnpm run test:size: passed, 16.37 kB / 16.5 kB.

Note: on Windows, raw pnpm test fails before running because the package script uses POSIX FORCE_COLOR=1; I ran the equivalent test:* scripts individually with $env:FORCE_COLOR=1.