feat: --benchmark mode (pure-Python hashrate baseline)

[devAsmodeus]

Summary

Adds hope-hash --benchmark for measuring the pure-Python SHA-256 baseline
before any C/Rust/SIMD/GPU optimizations (ROADMAP levels 2–3). Without a
"before" number, "after" numbers mean nothing.

What it does

• Reuses the existing parallel.start_pool machinery so the measured code
  path is exactly the same as real mining (including mid-state SHA-256
  from v0.3.0).
• Synthetic header (same shape as real block headers) + target=0 so
  workers never find a hit and just keep hashing for the full duration.
• Reports platform, Python build, CPU, total hashes, wall time, total H/s,
  per-worker H/s.
• Returns BenchResult dataclass for programmatic use (future scaling
  benchmarks, JSON exports).

Sample run

Intel i7-12700H, 4 workers, 5 seconds:

[bench] platform: Windows-10-10.0.26200-SP0
[bench] python:   3.11.9 (cpython)
[bench] cpu:      16 logical cores
[bench]   t=  1.0s  hashes=     2,654,208  rate=2.64 MH/s
[bench]   t=  3.0s  hashes=     9,093,120  rate=3.01 MH/s
[bench] === result ===
[bench]   total hashes:   11,436,032
[bench]   wall time:           5.01s
[bench]   hashrate:        2.28 MH/s
[bench]   per-worker:    570.23 KH/s (workers: 4)

This number (~570 KH/s/worker on a modern x86 core with hashlib) is the
target to beat with the next optimization step.

Why this approach

A few alternatives considered and rejected:

• C-extension via ctypes/cffi: marginal gain over hashlib (which already
  wraps OpenSSL/system libcrypto); the real Python overhead is between hashes
  (struct.pack, copy, int comparison), not in the hash itself. Skip until we
  rewrite the whole inner loop in C.
• Rust via PyO3: the right next step (the sha2 crate has SIMD baked
  in), but breaks the pure-stdlib invariant — needs an explicit decision
  before we go there. This PR keeps that decision separate.

Test plan

• py -3.11 -m unittest discover -s tests -v — 101/101 pass locally
• hope-hash --benchmark --bench-duration 5 --workers 4 runs cleanly
  and reports MH/s
• hope-hash --benchmark --demo rejected with explicit error
• CI matrix Python 3.11/3.12/3.13 × ubuntu/windows/macos

Related

Sets up the baseline measurement requested by ROADMAP "Идеи поверх скелета /
Бенчмарк-режим" and unblocks the C/Rust/SIMD discussion in levels 2–3.

🤖 Generated with Claude Code