ci(playwright): give Nextcloud a cache backend — there was none at all

[rubenvdlinde]

There was no cache backend at all

This workflow is 7k lines and configures none: no memcache.*, and no cache
extension in any setup-php. Nextcloud has no cache unless one is configured,
so ICacheFactory::createDistributed() handed every caller a null cache and
nothing survived from one HTTP request to the next.

No real instance runs that way, and it silently changes the behaviour of
anything session- or cache-backed.

Measured on filinq

OpenRegister's McpProtocolService keeps MCP sessions in
$cacheFactory->createDistributed(...). Every JSON-RPC call therefore arrived
without its session, and the whole agent-document-editing suite (6 tests)
stood down with:

the MCP session does not survive between requests on this instance …
OpenRegister stores sessions in a cache and this harness has no cache
backend. These specs pass against a real instance.

That skip was correct about the environment. The environment was the defect.

What this adds

apcu in the Playwright job's extensions	plus ini-values: apc.enable_cli=1 — php -S is the CLI SAPI, and without that flag APCu loads but stays inert, which looks exactly like never having installed it
a Configure the Nextcloud cache backend step	right after maintenance:install, before the app under test is enabled; sets memcache.local and memcache.distributed

APCu for both tiers — right here, wrong in production: this job is a single
php -S process tree on one runner, so "local" and "distributed" are the same
memory, and forked workers inherit APCu's shared segment.

The step fails loudly if apcu is missing or disabled rather than proceeding
without a cache, because a silently absent cache is the exact failure being
fixed. Guard verified in both directions locally: the apcu form exits 1 where
apcu is absent, and the same shape exits 0 for an extension that is present.

Scope

Deliberately the Playwright job only. The newman and phpunit jobs have the
same gap and are left for a follow-up — this is the one with a measured
consequence, and a shared workflow that all 18 fleet apps run should change one
job at a time.

YAML re-parsed after the edit: the new step is index 6 of 32, between "Install
Nextcloud" and "Cache Composer downloads".

Surfaced by the e2e skip-discipline gate, #559.