Speed up smoke tests: cache works_df, combine aggregations, multi-node cluster

- Cache works snapshot DataFrame once and reuse across all 9 test suites
  instead of re-reading the 300M+ row table in every cell
- Combine Tests 5 and 5b into a single aggregation pass (12 conditional
  counts in one scan instead of ~10 separate scans)
- Combine Test 7 null checks into a single aggregation
- Switch cluster from single-node i3.4xlarge to 4-worker i3.2xlarge
  for parallel reads

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: caseydm

jobs/snapshot_full.yaml

@@ -326,12 +326,11 @@ resources:
               availability: SPOT_WITH_FALLBACK
               zone_id: auto
               spot_bid_price_percent: 100
-            node_type_id: i3.4xlarge
+            node_type_id: i3.2xlarge
+            num_workers: 4
             enable_elastic_disk: false
             data_security_mode: SINGLE_USER
             runtime_engine: STANDARD
-            kind: CLASSIC_PREVIEW
-            is_single_node: true
       git_source:
         git_url: https://github.com/ourresearch/openalex-walden.git
         git_provider: gitHub