docs: rewrite README for investor narrative + refresh to May 2026 models (#22)

This rewrites the README around an investor-readable narrative:

- New top-of-page pitch: "take a 3 B-active open-weight model, close most
  of the SWE-bench Verified gap to GPT-5.5, overnight, hands-off, fully
  audited". Replaces the GSM8K math-tutor scenario, which was easy to
  read as "a small model doing small problems".

- New "30-second pitch" + "Why this matters" section before the
  technical content, so non-technical readers reach the value
  proposition (economics / engineering / strategy) within 30 seconds.

- "Our v1.1 target" replaces the prior "See it in action" section. The
  example is now framed explicitly as a roadmap target, not a logged
  run, to remove any read-as-claimed-result risk.

- Refreshed all model references to May 2026 reality:
    Baseline:  Qwen3.6-35B-A3B (3 B active, Apache 2.0, released
               2026-04-16, 73.4 % SWE-bench Verified — publicly
               reported)
    Frontier:  GPT-5.5 (88.7 %, OpenAI 2026-04-23),
               Claude Opus 4.7 (87.6 %)
    Also:      Gemini 3.1 Pro, Kimi K2.6 references
  All prior 2024-era model references (DeepSeek-Coder-V2-Lite, GPT-4o,
  GPT-4.1, Claude 3.5 Sonnet, Qwen2.5-Coder-32B) removed.

- New "Known limitations" section: honestly states what v1.0 cannot do
  (LLM evaluator can be fooled, sandbox is filesystem-only not network,
  history is summary not replay, cost can compound, real-provider
  integration not in CI).

- Fixed two stale numbers carried over from v0.2:
    "~1,200-line runtime" → "~1,900-line runtime"  (actual wc -l)
    "39 tests" → "99 tests"  (actual current count)

- Both English (README.md) and Chinese (README.zh.md) updated in lock-step.

No code changes. Test suite still 99 / 99 green.

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

Author: Protocol-zero-0