ci: make the CUDA job able to fail, and able to run (#144)

Two independent defects in the CUDA syntax job, both caught on its first run.

1. `cuda: '12.6.3'` — Jimver/cuda-toolkit@v0.2.19's version table stops at
   12.6.2, so the install step died with "Version not available: 12.6.3"
   before nvcc was ever invoked. One digit.

2. `nvcc ... 2>&1 | head -40` — a pipeline exits with the status of its LAST
   command, so head's 0 masked every nvcc error. The job printed "CUDA syntax
   check passed" unconditionally: it could not fail. Defect 1 is why we found
   out, since it broke the step *before* the pipe.

The second one is the one that matters. backend_cuda.cu is compiled by nothing
else in this repo — no local build, no test — so this job is the only thing
standing between a CUDA change and a user's GPU. A check that cannot fail is
worse than no check: it buys false confidence in exactly the file that most
needs the real thing. #298 spent a night debugging CUDA against no oracle at
all; this job is supposed to be that oracle.

It is also, precisely, the disease of the week in YAML form: a signal that
measures its own intention rather than the thing it claims to measure. See the
`~335 GB I/O saved` counter that counted dropped experts instead of bytes not
read (#303), and `route_agree: 95.3%` cited as "quality preserved" when it
only measures which experts coincide.

The other three jobs (engine, web, python) passed on the first run.

Co-Authored-By: ZacharyZcR <#144>

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -25,14 +25,20 @@ jobs:
       - name: Install CUDA toolkit (compiler only)
         uses: Jimver/cuda-toolkit@v0.2.19
         with:
-          cuda: '12.6.3'
+          # v0.2.19's version table stops at 12.6.2 — 12.6.3 fails the install
+          # step with "Version not available" before nvcc is ever reached.
+          cuda: '12.6.2'
           method: network
           sub-packages: '["nvcc"]'
       - name: Compile backend_cuda.cu (syntax only, no GPU)
         run: |
           cd c
+          # NOT `nvcc ... | head -40`: a pipeline exits with the status of its
+          # LAST command, so head's 0 masked every nvcc error and the job could
+          # only ever be green. A check that cannot fail is worse than no check —
+          # it buys false confidence in the one file nothing else compiles.
           nvcc -O2 -std=c++17 -arch=sm_80 -c backend_cuda.cu -o /dev/null \
-            -Xcompiler=-Wall,-Wextra 2>&1 | head -40
+            -Xcompiler=-Wall,-Wextra
           echo "CUDA syntax check passed"
 
   web: