ci_mirror: build the clean venv with uv (18.3 s -> 2.1 s cold)

The venv rebuild after a requirements.txt change was the largest cold-path
cost of `scripts/check_gate_stamp.sh full`. uv is also what CI installs
with, so the mirror gets closer to CI, not further. The interpreter is
passed as a path: `--python python3` let uv pick an active ~/.venv (3.13)
that crdt-sync 0.9.2 (>=3.14) cannot install into. pip stays as fallback.

Measured under capped.sh (4G, 20 %): `full` warm 10.9 s; fully cold
(no venv, empty PRE_COMMIT_HOME, empty mypy cache) 45.9 s, floor = the
pylint hook env install (~24 s, only after .pre-commit-config.yaml changes).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xfj1sxKH9VBPqpcAni8Jry

scripts/ci_mirror.sh

@@ -64,6 +64,20 @@ ensure_venv() {
 
     log "requirements.txt changed — rebuilding clean venv (mirrors CI install)"
     rm -rf "$VENV_DIR"
+    # uv is what CI installs with (`uv pip install -r`), and it rebuilt this
+    # venv in 2.1 s where venv + pip took 18.3 s (2026-10-16) -- the largest
+    # cold-path cost of `check_gate_stamp.sh full`. pip stays as the fallback.
+    if command -v uv >/dev/null 2>&1; then
+        # A PATH, not the bare name: `--python python3` let uv prefer an
+        # active ~/.venv (3.13) over the python3 the pip path would use.
+        uv venv --quiet --python "$(command -v python3)" "$VENV_DIR" \
+            || fail "uv venv for the clean venv"
+        uv pip install --quiet --python "$VENV_DIR/bin/python" -r "$REQ_PATH" \
+            || fail "uv pip install -r $REQUIREMENTS_FILE (a dep may be undeclared)"
+        printf '%s' "$current" > "$HASH_FILE"
+        log "clean venv ready (uv)"
+        return
+    fi
     python3 -m venv "$VENV_DIR"
     "$VENV_DIR/bin/python" -m pip install --quiet --upgrade pip \
         || fail "pip self-upgrade in the clean venv"