Add missing Sedona node counts (2, 12, 16) to benchmark matrix

[jathavaan]

Problem

The benchmark matrix for Sedona experiments is incomplete. Only 4 and 8 nodes exist at small scale, and 12-node clusters are missing entirely. The intended matrix is 2, 4, 8, 12, 16 nodes × all dataset sizes for both broadcast and partitioned strategies.

Target matrix

Broadcast join

Nodes	Small	Medium	Large
2	new	exists	exists
4	exists	exists	exists
8	exists	exists	exists
12	new	new	new
16	new	new	exists

Partitioned join

Nodes	Small	Medium	Large
2	new	skip:failed	skip:failed
4	exists	skip:failed	skip:failed
8	exists	skip:failed	skip:failed
12	new	skip:failed new	skip:failed new
16	new	skip:failed new	skip:failed new

All partitioned medium/large entries are skip: failed (executor OOM on RangeJoin spatial index).

Required changes per new entry

1. Entry in benchmarks.yml with id, image, cpu, memory_gb, dataset_size, runs, related_script_ids
2. Docker image for 12-node variants (new image tag)
3. Entrypoint in src/presentation/entrypoints/ for 12-node variants
4. case arm in benchmark_runner.py for 12-node variants
5. Batch pairing must satisfy the four constraints (same query type, same dataset_size, at most one PostGIS, Databricks vCPU sum ≤ 200)