ci: labeler workflow uses fork's own labeler.yml, not upstream

[Marxist-Leninist]

The labeler workflow hard-codes repository: ggml-org/llama.cpp on its checkout step, so it always reads upstream's labeler.yml. Upstream uses v5 all: composition syntax which actions/labeler@v6 rejects:

Error: Unknown config options were under "changed-files": all

Every PR to this fork has been failing the labeler check since the action was bumped to v6.

Drop the hard-coded repository: so the checkout uses its default (this fork's master), which already has correct v6 syntax. Decouples the fork from upstream's labeler.yml schema.

[Marxist-Leninist]

Heads up: this PR's own labeler check is also failing, and that's expected — it's a GitHub Actions catch-22 with pull_request_target.

pull_request_target events run the workflow from the base branch, not the PR head. That's a security feature to prevent untrusted PRs from modifying their own CI to abuse write permissions. So when this PR runs, GitHub runs the current master's labeler workflow (the broken one), not the fixed version in this PR. The fix can't fix its own check — it only takes effect on PRs created after this one is merged.

Two ways forward:

1. Merge this PR despite the failing labeler check. It's informational only, doesn't block merge, and the failure is literally the thing this PR fixes. Once merged, all future PRs to PrismML-Eng/llama.cpp will get a green labeler check.

2. Wait for the upstream fix at ci: fix labeler.yml v6 syntax — drop all: composition ggml-org/llama.cpp#21627. That flattens upstream's all: composition syntax. Because the current PrismML-Eng workflow checks out upstream's labeler.yml at runtime, fixing it upstream will make every existing failing check on every fork start passing on next run with no further action — including this PR's check, which would then turn green and allow normal merge.

Either way works. Option 1 is faster and makes this fork self-sufficient (no longer dependent on upstream's labeler.yml schema being v6-compatible).

[khosravipasha]

This is only for the CI and actions?
I think its okay, long term we plan to merge all our good backends to llama.cpp so this fork would mostly become archival or for experimentation.
I will pull the upstream changes later.