docs(prompts): survey agent prompts for general capability improvements

[minixalpha]

This PR adds a bilingual study of agent system prompts, motivated by benchmark runs that completed normally but failed independent verification. It proposes guidance on requirement coverage, independent checks, failure diagnosis, and evidence-based completion, with a controlled evaluation plan. The candidate remains research text; runtime prompts are unchanged and no new benchmark results or capability gains are claimed.

The archive contains 42 English originals and full Chinese translations across Codex, OpenCode, Pi, Grok Build, DeepSeek Harness, nanoPyCodeAgent, and three user-provided Claude snapshots. It includes bilingual analyses and indexes, pinned source revisions, extraction selectors, original/translation hashes, and upstream licenses. Templates, builders, test fixtures, and chat-product prompts are distinguished from complete coding-agent runtime prompts.

Validation:

• All 39 local-source originals match their pinned sources; all 42 original/translation hash pairs pass.
• The three user-provided Claude bodies remain unchanged, with all 183 paragraphs covered and tags, placeholders, model identifiers, URLs, and fences preserved.
• All 469 local Markdown links resolve, and fences are balanced. The 33 whitespace warnings are confined to archived prompt blocks and preserved for fidelity.
• CI passes on 3944747: pytest on Python 3.13 and 3.14, plus Harbor workflow contracts.

Documentation status: all 44 Chinese research files have synchronized English counterparts, and the English 0.10.x development notes are synchronized with the Chinese source. Both root READMEs are unchanged. No changelog entry is added because this PR documents internal research without changing user-facing behavior.