CI: php.yml (Valgrind + test matrix) not triggered on valkey-glide submodule bumps due to paths filter

[prateek-kumar-improving]

Summary

The PHP CI workflow (.github/workflows/php.yml), which contains the
Valgrind job and the full PHP test matrix, is gated by a paths: filter on
both push and pull_request. That filter does not match a submodule-pointer
bump of valkey-glide, nor the top-level utils/ tooling. As a result, PRs
that update the glide-core submodule can merge without the Valgrind job (or the
php.yml matrix) ever running, letting glide-core regressions slip in
unexercised.

Evidence

PR #325 ("Update submodule glide core") changed only:

• utils/patch_proto_and_rust.py
• valkey-glide (submodule pointer bump to the Valkey 9.0 core)

gh pr checks 325 shows no Valgrind job ran on that PR. The checks that did
run/pass came from other workflows; php.yml was skipped entirely.

The Valgrind leak-gate failures now tracked in #328 (glide-core global
possibly lost allocations) were introduced with that 9.0 core bump but only
surfaced later on PR #321, which is the first PR since the bump to touch a
php.yml-triggering path (*.c / *.h / *.php / tests/**).

Root cause (path filter)

php.yml watches specific submodule subpaths, e.g.:

valkey-glide/glide-core/src/**
valkey-glide/glide-core/redis-rs/redis/src/**
valkey-glide/ffi/src/**
valkey-glide/utils/cluster_manager.py

A submodule-pointer update is recorded as a change to the valkey-glide gitlink
(a single tree entry), which does not expand into the individual file paths
inside the new submodule commit. So none of the valkey-glide/.../** globs
match, and the workflow is skipped. Likewise, the top-level utils/** directory
is not in the trigger list (only valkey-glide/utils/cluster_manager.py is).

Proposed fix

• Add trigger paths so glide-core bumps and tooling changes run php.yml, e.g.:
  • valkey-glide (the submodule gitlink itself)
  • utils/**
• Alternatively, treat submodule updates as always-trigger for php.yml (or run
  the Valgrind/test matrix on a schedule against main as a safety net).
• Verify by opening a test PR that only bumps the submodule and confirming the
  Valgrind job is triggered.

Priority

This should be addressed before #328: without CI actually running the
Valgrind job on submodule bumps, regressions like the one in #328 will continue
to land undetected regardless of any suppression work.

Acceptance criteria

• A submodule-only bump of valkey-glide triggers php.yml (Valgrind +
  test matrix).
• Top-level utils/** changes trigger php.yml.
• Confirmed with a test PR that the Valgrind job runs.