perf(ci): 8 vCPU for ns7zip build VM (4 was starving parallel build)

WSL on the host builds the 3 configs in parallel in ~103s (= one
config) — the parallel design is correct. The 4-vCPU CI VM ran 3x
make -j on 4 cores -> oversubscription thrash -> ~3x slower. Not
structural / not RAM. Bump guest-cpu to 8 (job.py patches the clone
VMX numvcpus); host is 10c/20t, this matrix is 2 VMs.

.gitea/workflows/build-ns7zip.yml

@@ -46,9 +46,11 @@ jobs:
           # Cross-repo build: github.ref_name is this repo's branch, not
           # nsis-plugin-ns7zip's. Pin the target repo's branch.
           repo-branch:     'main'
-          # 8 GB: the parallel MinGW cross-compile thrashes at the 4 GB
-          # Linux template default (3 concurrent make -j). Harmless for
-          # Windows (host has ample RAM).
+          # 8 vCPU / 8 GB. WSL on the host builds the 3 configs in
+          # parallel in ~103s (= one config); the 4-vCPU CI VM thrashes
+          # (3x make -j on 4 cores) and takes ~3x. Not structural — just
+          # vCPU starvation. Host is 10c/20t; this matrix = 2 VMs.
+          guest-cpu:       '8'
           guest-memory-mb: '8192'
           # Suffix disambiguates artifact dirs: F:\CI\Artifacts\{run_id}-{attempt}-windows
           job-id-suffix:   '${{ matrix.target }}'