Lexer: extract the matches with PREG_PATTERN_ORDER

PREG_SET_ORDER allocates one array per match. Now that the token patterns
have no capturing groups those arrays hold two entries each -- the whole
match and the MARK -- and tokenize() reads both out and drops the array
again, so lexing pays one array allocation per token for nothing.

PREG_PATTERN_ORDER collects the whole PHPDoc into two arrays instead:
$matches[0] holds the values and $matches['MARK'] the token types. Two array
headers per PHPDoc rather than one per token.

Measured over 37150 unique docblocks (8.38 MiB, 2.36M tokens) collected from
20 OSS libraries, one full vendor tree, phpstan-src and this library:

              lexing   parse (lex + PhpDocParser)
  PHP 8.1      -42%
  PHP 8.4      -40%
  PHP 8.5      -40%    -19%

tokenize() still returns list<array{string, int, int}>, and both the tokens
and the ASTs printed from them are byte-identical to before over the whole
corpus.

Author: JanTvrdik

src/Lexer/Lexer.php

@@ -5,7 +5,6 @@
 use PHPStan\PhpDocParser\ParserConfig;
 use function implode;
 use function preg_match_all;
-use const PREG_SET_ORDER;
 
 /**
  * Implementation based on Nette Tokenizer (New BSD License; https://github.com/nette/tokenizer)
@@ -116,13 +115,24 @@ public function tokenize(string $s): array
 			$this->regexp = $this->generateRegexp();
 		}
 
-		preg_match_all($this->regexp, $s, $matches, PREG_SET_ORDER);
+		// PREG_PATTERN_ORDER, not PREG_SET_ORDER: it collects the whole PHPDoc
+		// into two arrays instead of allocating one array per token. This only
+		// pays off while the token patterns have no capturing groups, because
+		// every group would get an array of its own, one entry per token.
+		preg_match_all($this->regexp, $s, $matches);
+
+		$values = $matches[0];
+		if ($values === []) {
+			return [['', self::TOKEN_END, 1]];
+		}
+
+		$marks = $matches['MARK'];
 
 		$tokens = [];
 		$line = 1;
-		foreach ($matches as $match) {
-			$type = (int) $match['MARK'];
-			$tokens[] = [$match[0], $type, $line];
+		foreach ($values as $i => $value) {
+			$type = (int) $marks[$i];
+			$tokens[] = [$value, $type, $line];
 			if ($type !== self::TOKEN_PHPDOC_EOL) {
 				continue;
 			}
@@ -137,6 +147,7 @@ public function tokenize(string $s): array
 
 	private function generateRegexp(): string
 	{
+		// every group in here must be non-capturing, see tokenize()
 		$patterns = [
 			self::TOKEN_HORIZONTAL_WS => '[\\x09\\x20]++',