Use 32GB runner and full concurrency for CLI builds

The build-cli and build-cli-pr jobs were OOMing on the 16GB runner
when running 3 concurrent deno compile processes simultaneously.
Each cross-compilation target loads ~500–600MB of source files, and
peak memory during V8 snapshot generation pushes well beyond 16GB
with 3 concurrent processes.

Upgrade both jobs to the 32GB runner profile and increase CONCURRENCY
from 3 to 5 (matching the total number of cross-compilation targets)
so all 5 targets build in a single parallel round.

Assisted-by: Claude Code:claude-sonnet-4-6

Author: dahlia

.github/workflows/main.yaml

@@ -265,15 +265,15 @@ jobs:
   build-cli-pr:
     if: github.event_name == 'pull_request'
     needs: [release-test]
-    runs-on: namespace-profile-linux-amd64-16gb
+    runs-on: namespace-profile-linux-amd64-32gb
     steps:
     - uses: actions/checkout@v4
     - uses: ./.github/actions/setup-mise
     - run: mise run codegen
     - run: deno task pack
       working-directory: ${{ github.workspace }}/packages/cli/
       env:
-        CONCURRENCY: "3"
+        CONCURRENCY: "5"
     - uses: actions/upload-artifact@v4
       with:
         name: cli-pr-${{ github.event.pull_request.number }}
@@ -324,7 +324,7 @@ jobs:
   build-cli:
     if: github.event_name == 'push'
     needs: [test, test-node, test-bun, test-cfworkers, lint, release-test, determine-version]
-    runs-on: namespace-profile-linux-amd64-16gb
+    runs-on: namespace-profile-linux-amd64-32gb
     steps:
     - uses: actions/checkout@v4
     - uses: ./.github/actions/setup-mise
@@ -333,7 +333,7 @@ jobs:
     - run: deno task pack
       working-directory: ${{ github.workspace }}/packages/cli/
       env:
-        CONCURRENCY: "3"
+        CONCURRENCY: "5"
     - uses: actions/upload-artifact@v4
       with:
         name: cli