Generalize presto-nvl72 slurm scripts for any cluster

[misiugodfrey]

Summary

• Make presto/slurm/presto-nvl72/ cluster-agnostic: all per-cluster values (partition, account, cpus-per-task, time limits, images, image paths, data root, etc.) now come from ~/.cluster_config.env. See cluster_config.env.example.
• Introduce two shared libraries: launcher_common.sh (cluster-variant resolution, sbatch arg assembly, preflight checks, job monitoring) and slurm_common.sh (shared .slurm preamble). Eliminates ~80% of the per-launcher duplication.
• Pre-flight checks in every launcher: missing image / data dir / analyzed metastore produce actionable errors pointing at the exact command to run next, before any sbatch submission.
• Job state surfaced via sacct: launcher prints Job FAILED (state: …, exit: …) and always displays stderr, so silent failures don't masquerade as success.
• launch-analyze-tables.sh is now always CPU (ANALYZE TABLE disables cudf regardless). New CLUSTER_DEFAULT_VARIANT setting lets CPU-only clusters drop the --cpu flag.
• gen-data uses the same run_py_script.sh + miniforge3 pattern as analyze/benchmark instead of inventing its own pip flow.
• Several CPU-variant fixes in functions.sh:run_worker (gate NVIDIA_VISIBLE_DEVICES export and GDS bind-mounts on VARIANT_TYPE=gpu so CPU nodes don't hit enroot hook failures).
• README rewritten as a three-step workflow walkthrough; all 6 entry points now documented.

Test plan

• ./launch-gen-data.sh -s 1 on CPU cluster → 240 MB of parquet under ${DATA}/tpch-rs-1
• ./launch-analyze-tables.sh -s 1 on CPU cluster → tpchsf1/ populated in .hive_metastore/
• ./launch-run.sh -n 1 -s 1 --cpu on CPU cluster → all 22 TPC-H queries pass
• ./launch-run.sh -n 1 -s 1 on GPU cluster (job currently pending in batch queue)
• download image -> gen data -> analyze data -> run_benchmark for tpch sf1k on NVL72 GPU/CPU
• Preflight failure modes: missing image / missing data / missing metastore each print correct actionable hint
• Failure reporting: jobs that exit non-zero are surfaced as FAILED with stderr displayed

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[kingcrimsontianyu]

I would suggest keeping these two simple examples in Step 3 — Run benchmarks, minus the -w -c arguments.

[kingcrimsontianyu]

For the "test plan", I'd suggest checking if --disable-gds, -p, --nsys-worker-id, -q <query_list> continue to work on the GPU cluster.