[TwigComponent] Fix quadratic scanning in TwigPreLexer::consume()

[Kocal]

Q	A
Bug fix?	no
New feature?	no
Deprecations?	no
Documentation?	no
Issues	-
License	MIT

consume() built a copy of the whole remaining template on every call via
substr(), and it is called several times per character in the main scan
loop. Pre-lexing was therefore quadratic in template size.

Reuse check(), which compares in place with substr_compare().

Pre-lexing a 154 KB template goes from ~724 ms to ~94 ms, and scaling is
now linear.

Benchmarked from the repository root with blackfire run symfony php bench.php:

<?php
require __DIR__.'/src/TwigComponent/vendor/autoload.php';

use Symfony\UX\TwigComponent\Twig\TwigPreLexer;

$chunk = <<<'TWIG'
<div class="card">
    <h1>Some title here</h1>
    <p>Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.</p>
    {% if foo %}
        <span>{{ bar|upper }}</span>
    {% endif %}
    <twig:Alert type="success" :count="items|length">
        Hello world
    </twig:Alert>
</div>

TWIG;

$input = str_repeat($chunk, 500); // ~154 KB
(new TwigPreLexer())->preLexComponents($input);

Blackfire:

• before — 3.34s wall / 3.18s CPU: https://blackfire.io/profiles/8ef533c9-e96a-4b48-8f04-2711c8749ee6/graph
• after — 2.22s wall / 2.19s CPU: https://blackfire.io/profiles/a7ccabce-35d6-4163-988d-27ca37a439b4/graph
• diff: https://app.blackfire.io/profiles/compare/8ef533c9-e96a-4b48-8f04-2711c8749ee6...a7ccabce-35d6-4163-988d-27ca37a439b4/graph

Analysis, implementation and benchmarks by Claude Opus 5.

[smnandre]

I'm a bit 😄 at the title .. compared to the IRL effects*, but this goes into the good direction so 👍

(* in production: not a nanosecond of difference, as templates are pre-compiled once during deploy, the average template is probably more around 5kb, and I doubt even 5% of them would even be concerned by this case.)