Integrate torc for benchmark orchestration (optional feature)

[pesap]

Description

Add optional torc integration to `arco-bench` for distributed benchmark orchestration. This enables:

• Parallel benchmark execution across multiple nodes
• SQLite-based result storage with full provenance (git commit, hardware info, timestamps)
• Change detection (only re-run benchmarks for changed code)
• REST API to query historical performance data
• Slurm integration for HPC benchmark runs

Implementation

Add `torc` feature flag to `arco-bench`:

When `torc` feature enabled:

• Benchmarks defined as KDL workflow files (similar to torc workflows)
• Submit to torc server via HTTP API
• Poll for completion with resource metrics
• Store results in torc's SQLite database

When `torc` feature disabled (default):

• Run benchmarks locally with current behavior
• JSONL output for CI comparison

Usage

Context

Depends on #139 (Locked benchmark suite). The benchmark suite should be implemented first with local execution, then this adds the optional torc layer for distributed/tracked runs.

Acceptance Criteria

• `arco-bench` compiles with and without `torc` feature
• Benchmark workflows can be defined in KDL format
• CI job demonstrates torc-based benchmark execution
• Documentation for setting up local torc server for benchmarks

Target: v0.5 (optional enhancement to benchmark suite)
Priority: Medium (blocks on #139)

[pesap]

Progress Update

Completed in PR #160:

✅ Feature flag setup ( feature with conditional compilation)
✅ Runner module structure (torc/local backends)
✅ Benchmark workflow KDL format (examples/benchmark-workflow.kdl)
✅ API research documented (/tmp/torc-api-summary.md)

Next tasks for follow-up:

• HTTP client implementation (torc API client)
• Result adapter (convert torc job results to JSONL)
• CI integration with GitHub Actions + torc server

The foundation is laid. Remaining work can proceed in parallel once this PR merges.