Add per-phase heap profiler

[gabrielbosio]

Adds heap profiling infrastructure behind the existing instruments + jemalloc-stats feature flags.

• scripts/bench_heap_profile.sh runs the prover across program sizes and prints a per-phase heap breakdown with growth rates and extrapolation
• stark::instruments gains set_heap_reader, snap, HeapSnapshot, ProveHeapProfile for phase-boundary heap readings
• prover::instruments::print_report prints a === HEAP PROFILE (MB) === table when a heap reader is registered

Usage: scripts/bench_heap_profile.sh --programs "1M 2M 4M 8M"

[gabrielbosio]

/bench

[github-actions]

Benchmark — fib_iterative_8M (median of 3)

_{Table parallelism: 32 (auto = cores / 3)}

Metric	main	PR	Δ
Peak heap	250570 MB	250314 MB	-256 MB (-0.1%) ⚪
Prove time	81.704s	82.235s	+0.531s (+0.6%) ⚪

✅ No significant change.

✅ Low variance (time: 0.8%, heap: 0.3%)

_{Commit: 0ff1b1b · Baseline: cached · Runner: self-hosted bench}

[gabrielbosio]

Superseded by #507