Quote non-token characters in matchFormatPattern

[iliaal]

hasFormatWithModifiers() interpolated the format string into the final PCRE pattern without quoting it. Only format letters and modifiers were replaced; every other byte (parentheses, quantifiers, alternation, anchors) kept its regex meaning, so a caller-supplied format could alter the structure of the validation pattern rather than match literally:

Carbon::hasFormatWithModifiers('kkkk', 'k{4}'); // true before this PR: '{4}' acted as a quantifier
Carbon::hasFormatWithModifiers('qqqq', 'q{4}'); // true before this PR
Carbon::hasFormat('qqqq', 'q{4}');              // false: the sibling API preg_quotes its input

The inconsistency with hasFormat() also meant the two validators disagreed on identical inputs.

How it is fixed

matchFormatPattern() now wraps each token-generated regex fragment between null-byte sentinels during the substitution pass, then quotes every remaining byte of the pattern. Backslash escapes in the format (\T, \.) still resolve to their literal target, and token fragments (d, #, *, ?, !, |, +) keep their regex meaning. Raw null bytes in the format are stripped before processing so they cannot collide with the sentinels.

Behavior change to be aware of: a format containing unescaped regex metacharacters that used to be interpreted as pattern syntax (e.g. .*, (a|b)) will now only match those characters literally. This matches what hasFormat() already did and what createFromFormat() escape semantics imply.

[kylekatarnls]

I confirm that it was not initially intended that {4} would be read as a multiplier.

The "modifier" in this context refers to the non-date characters listed in https://www.php.net/manual/en/datetimeimmutable.createfromformat.php#datetimeimmutable.createfromformat.parameters

However I suspect some might have misinterpreted this as a feature, may have think that {number} is one of the supported modifiers.

I'll add a warning in the release since it can be a breaking change for a minority of users.

I still proceed as I think the number of use cases where a format would contain multiple in a row the same unit must be quite small.