Add CacheOptionalInterface to more blocks

Added it to powered, messages and also page title at first but reverted that again because page title is already not cached further up, so this has no meaning there, we never get so far.

We'll want to review render times vs cache load times for this. I did some tests before around this. Performance tests are of course "happy", as in fewer cache lookups. But they don't measure if it's actually a good thing in terms of memory/cpu, at least not as assertions.

Tempted to won't fix this. I profiled this in two ways, and both show that the cached blocks are actually faster even for something as basic as the powered by block. what I didn't account for is that each block also includes a block template and the full twig rendering around that.

I tested this with the following script that I executed with drush scr:

use Drupal\block\Entity\Block;

$block = Block::load($extra[0]);

$renderer = \Drupal::service('renderer');
$view_builder = \Drupal::entityTypeManager()->getViewBuilder('block');
foreach(range(1, 1000) as $i) {
  $build = $view_builder->view($block);
  $renderer->renderPlain($build);
}

With xhprof, Drupal\Core\Render\Renderer::doRender takes around 420ms for a cached block such as olivero_site_branding. and 650ms for a non-cached block (olivero_powered). But the profiling overhead needs to be considered.

I also tried out https://github.com/sharkdp/hyperfine, which is why I did this as a CLI script. Saw a video about that recently, it's pretty neat as it can compare multiple runs with parameters, automatically calculates averages and accounts for variation. That gave me this output:

Benchmark 1: ddev drush scr local/cached_block.php -- olivero_powered
  Time (mean ± σ):     591.7 ms ±  12.6 ms    [User: 79.6 ms, System: 18.2 ms]
  Range (min … max):   570.3 ms … 611.2 ms    10 runs

Benchmark 2: ddev drush scr local/cached_block.php -- olivero_messages
  Time (mean ± σ):     638.9 ms ±  16.4 ms    [User: 80.4 ms, System: 17.0 ms]
  Range (min … max):   619.9 ms … 669.6 ms    10 runs

Benchmark 3: ddev drush scr local/cached_block.php -- olivero_help
  Time (mean ± σ):     513.1 ms ±   6.6 ms    [User: 80.4 ms, System: 16.5 ms]
  Range (min … max):   499.8 ms … 523.3 ms    10 runs

Benchmark 4: ddev drush scr local/cached_block.php -- olivero_site_branding
  Time (mean ± σ):     514.2 ms ±  12.0 ms    [User: 78.3 ms, System: 17.8 ms]
  Range (min … max):   502.5 ms … 536.8 ms    10 runs

Summary
  ddev drush scr local/cached_block.php -- olivero_help ran
    1.00 ± 0.03 times faster than ddev drush scr local/cached_block.php -- olivero_site_branding
    1.15 ± 0.03 times faster than ddev drush scr local/cached_block.php -- olivero_powered
    1.25 ± 0.04 times faster than ddev drush scr local/cached_block.php -- olivero_messages

So, the two cached blocks are the same, which makes sense, and faster than powered. The script might not 100% reflect real world usage, the renderPlain() for example also already does the lazy builder within the messages block, which is why that is slower, but I don' think it's that far off.

What I did just notice is that I did run this with redis. I repeated it without redis and got this:

Benchmark 1: ddev drush scr local/cached_block.php -- olivero_powered
  Time (mean ± σ):     572.8 ms ±   7.9 ms    [User: 78.5 ms, System: 19.5 ms]
  Range (min … max):   562.8 ms … 586.4 ms    10 runs

Benchmark 2: ddev drush scr local/cached_block.php -- olivero_messages
  Time (mean ± σ):     626.2 ms ±  13.8 ms    [User: 79.4 ms, System: 17.7 ms]
  Range (min … max):   604.7 ms … 647.8 ms    10 runs

Benchmark 3: ddev drush scr local/cached_block.php -- olivero_help
  Time (mean ± σ):     557.0 ms ±   8.7 ms    [User: 79.6 ms, System: 15.8 ms]
  Range (min … max):   547.0 ms … 568.9 ms    10 runs

Benchmark 4: ddev drush scr local/cached_block.php -- olivero_site_branding
  Time (mean ± σ):     550.6 ms ±  13.6 ms    [User: 77.9 ms, System: 17.8 ms]
  Range (min … max):   532.7 ms … 575.9 ms    10 runs

Summary
  ddev drush scr local/cached_block.php -- olivero_site_branding ran
    1.01 ± 0.03 times faster than ddev drush scr local/cached_block.php -- olivero_help
    1.04 ± 0.03 times faster than ddev drush scr local/cached_block.php -- olivero_powered
    1.14 ± 0.04 times faster than ddev drush scr local/cached_block.php -- olivero_messages

In other words, it's actually redis (2.x) that gives the cached blocks the clear advantage, and it's pretty equal with the database backend, but I'm still not sure that makes sense then. That backs my results in https://www.md-systems.ch/en/blog/2025-09-12/performance-and-maintenance.... If the cache is slower than in my scenario with ddev, that might change things again? RedisBackend::get() only makes up 8% of the total time.

I also tested with asserts on and off and it didn't seem to make a meaningful difference.