docs(harness): add CLINE.md thin pointer and list Cline in AGENTS.md - #37

Summary\n\nCline joins Codex/Claude/Gemini as a listed harness with a thin entry file (CLINE.md) that delegates to AGENTS.md, matching the existing CODEX.md / CLAUDE.md / GEMINI.md shape. AGENTS.md is updated in two places to name Cline alongside the other harnesses, and to state explicitly that per-model entry files are intentionally not created — answering the recurring question of whether each new model needs its own .md.\n\n## Changes\n\n- CLINE.md (new) — thin harness entry pointer (23 lines) with Cline-specific notes: open Agent Skills discovery, no adapter-local manifest, and the empirically verified AGENTS.md auto-injection behavior.\n- AGENTS.md — added Cline to the harness enumeration (line 4) and CLINE.md to the entry-files paragraph (line 69), plus a paragraph stating per-model files (DEEPSEEK.md, GLM.md, KIMI.md, QWEN.md) are intentionally not created because those are model/API providers routed through a harness, not harnesses themselves.\n- tests/unit/test_cline_adapter.py — added test_cline_entry_file_is_thin_pointer_to_agents_md and test_agents_md_lists_cline_among_harnesses (5 cases total now).\n- changelog.md / handoff.md — one Unreleased line + handoff update.\n\n## Empirical verification of Cline ← AGENTS.md auto-injection\n\nRan an isolated Cline session (CLI 3.0.61, --data-dir $(mktemp -d), model cline-pass/glm-5.2) with the prompt: "Output ONLY the literal first line of this repository AGENTS.md file. Do not run any tools."\n\nResult: the model responded # AGENTS.md, and the reasoning trace explicitly said:\n\n> "Looking at the Workspace Configuration, I see the full content of the AGENTS.md file was provided in the workspace context."\n\nThis confirms Cline auto-injects the repo-root AGENTS.md into workspace context at session start — so AGENTS.md already reaches every model routed through ClinePass (DeepSeek, GLM, Kimi, Qwen, …) without any per-model plumbing. That is the empirical basis for the "per-model entry files are not created" policy recorded in AGENTS.md.\n\n## Why this shape\n\n- Harness entry files (CODEX.md, CLAUDE.md, GEMINI.md, now CLINE.md) are thin pointers because harnesses, not models, are the layer that owns project-context files. Cline auto-injection of AGENTS.md makes a CLINE.md body mostly a placeholder for Cline-specific quirks — same pattern as the other three.\n- Per-model files would be redundant: the harness owns context injection, so one harness entry covers every model the harness routes to.\n- This PR does not modify the harness-parity CI job — that is tracked separately as a Medium backlog item (PR #36).\n\n## Verification\n\n- tests/unit/test_cline_adapter.py: 5 cases (3 original + 2 new), re-run with Python 3.13 standalone (stdlib-only) — all pass.\n- git diff --stat: 5 files, +105/-40; Python scope limited to test_cline_adapter.py (stdlib pathlib only — no ruff/mypy surface beyond what already passed).\n\n## Examples Impact\n\n- [x] N/A — doc/test-only PR (docs:), no examples/ change required per PR template.\n\n## ADR Impact\n\n- [x] N/A — no architectural decision changed; this records an existing invariant (harness owns context injection) and empirically verifies it.\n\n## Local verification checklist\n\n- [x] git status clean before branch; branched off develop, base develop.\n- [x] Commit message follows Conventional Commits (docs(harness): ...).\n- [x] Unit tests re-run with Python 3.13 standalone — 5/5 pass.\n