feat: add --memory_profile_out flag for Chonk memory profiling

[johnathan79717]

Summary

• Adds --memory_profile_out <file> flag to bb prove that outputs a JSON report of memory consumption during Chonk proving
• Per-circuit polynomial memory breakdown by category (wires, sigmas, ids, selectors, tables, lookup, databus, etc.)
• RSS checkpoints at key stages (after polynomial allocation, after accumulate) to identify when peak occurs
• CRS (commitment key) memory tracking
• Python extraction script converts JSON into benchmark dashboard entries (stacked area chart for categories, line charts for totals)
• Integrated into ci_benchmark_ivc_flows.sh for continuous tracking

Refs: AztecProtocol/barretenberg#1641

Dashboard change needed

The extraction script uses extra: "stacked-area:..." which requires a new renderStackedAreaGraph function in benchmark-page-data/bench/index.html. The change is ready at /mnt/user-data/jonathan/benchmark-page-data on branch jh/stacked-area-chart and needs to be pushed by someone with access to AztecProtocol/benchmark-page-data. Until then, the memory metrics still appear as individual line charts (graceful fallback).

Example output

Peak RSS: 256 MiB at after_accumulate (circuit 6)
Peak circuit polynomial memory: 82.5 MB (circuit 6)
  selectors: 26.1 MB, wires: 14.9 MB, sigmas: 14.9 MB, ids: 14.9 MB
  other: 4.4 MB, z_perm: 3.7 MB, lookup: 3.3 MB, databus: 0.2 MB
CRS: 16.0 MB
Total circuits: 13

Test plan

• Builds cleanly (ninja bb, ninja chonk_tests, ninja ultra_honk_tests)
• Tested locally with deploy_ecdsar1+sponsored_fpc flow: JSON output is valid, all 13 circuits profiled with RSS checkpoints
• Extraction script produces correct stacked-area chart entries for benchmark dashboard
• CI build passes