refactor: simplify hex escape pattern scanning

[nkuprins]

Should be reviewed only once #1001 gets merged

Related: #1000

Purpose of the pull request

The second of the two PRs. This one does the refactor/cleanup. No behaviour change.

What's changed?

• The match loop is written once instead of twice. The scan used to live in two methods: escapeHex located the
  first valid pattern, then processWithPatterns located every later one. Both carried the same match-handling code - bounds check, closing-_ test, hex validation, append, so any change to the rule had to be made in two places. It is now a single loop.
• The no-match path allocates nothing. The StringBuilder is created on the first match rather than up front, so a cell containing no _xHHHH_ - costs one indexOf scan and no allocation.
• The offsets have names. Each iteration derives hexStart, suffixIndex and patternEnd from PREFIX_LENGTH and HEX_DIGIT_COUNT.
• isValidHexFast is now isHexDigits, and the lookup table has a note on why it is not Character.digit(c, 16) - that method also accepts non-ASCII digits such as U+0663, which OOXML never encodes.

Checklist

• I have read the Contributor Guide.
• I have written the necessary doc or comment.
• I have added the necessary unit tests and all cases have passed.