fix: handle stale .vllm-venv in test runner (#829)

* fix: handle stale .vllm-venv in test runner (#828)

Add --clear to the uv venv call so a leftover venv from a killed job
or incomplete cleanup is replaced instead of causing a hard failure.

Fixes #828

* fix: add .*-venv to .gitignore (#828)

When CACHE_DIR is unset the test runner creates .vllm-venv in the repo
root. Without a gitignore entry this multi-GB directory can be
accidentally committed or rsynced. Use a wildcard to cover any future
isolated venvs following the same naming pattern.

Author: planetf1

.gitignore

@@ -381,6 +381,7 @@ celerybeat.pid
 # Environments
 .env
 .venv
+.*-venv
 env/
 venv/
 ENV/

test/scripts/run_tests_with_ollama_and_vllm.sh

@@ -219,7 +219,7 @@ if [[ "$WITH_VLLM" == "1" ]]; then
             log "Reusing existing vLLM venv at $VLLM_VENV (KEEP_VLLM_VENV=1)"
         else
             log "Creating isolated vLLM venv at $VLLM_VENV ..."
-            uv venv "$VLLM_VENV" --python 3.11
+            uv venv "$VLLM_VENV" --python 3.11 --clear
             log "Installing vllm into $VLLM_VENV ..."
             uv pip install --python "$VLLM_VENV/bin/python" vllm \
                 > "$LOGDIR/vllm_install.log" 2>&1 \