docs(cli): spike chat and complete parity

[localai-org-maint-bot]

Outcome

Moves SERVE-CLI-CHAT from INVENTORIED to SPIKE with an accepted, CPU-only implementation design. No runtime command support is claimed by this PR.

The pinned-source audit found a material inventory bug: vLLM 5559679229 does ship both chat and complete through vllm/entrypoints/cli/openai.py. The prior matrix had stopped at the import list in main.py and incorrectly called them project extensions.

Design selected

The spike chooses a dual-mode command family:

• mirror vLLM remote chat and complete over OpenAI HTTP/SSE, including URL/auth/model discovery, quick and interactive modes, chat history, streaming, and TTFT/TPS
• preserve the shipped in-process vllm-cli --model DIR --prompt TEXT form as an explicit compatibility alias
• use the already-vendored httplib and nlohmann::json, with no new dependency
• split follow-on work into CPU-verifiable W1 parse/dispatch, W2 transport, W3 complete, W4 chat, and W5 packaging gates

Remote-only was rejected because it breaks the current library-first example. In-process-only was rejected because it does not mirror the pinned vLLM behavior.

Evidence

Pinned vLLM 5559679229bc961848b121ccdeaa8fa5d79bec98:

• command registration: vllm/entrypoints/cli/main.py:17-37,73-98
• client/model resolution and stream shaping: vllm/entrypoints/cli/openai.py:30-100
• chat: vllm/entrypoints/cli/openai.py:155-234
• complete: vllm/entrypoints/cli/openai.py:237-312

Local baseline:

• in-process completion only: examples/cli/main.cpp:1-207
• C-ABI stream semantics: tests/capi/test_capi.cpp:567-711
• chat-template baseline: tests/capi/test_chat_prompt.cpp:37-89

Verification

All CPU-only gates pass on commit 73566588538e4956141be640a9a1603280785f11:

• python3 scripts/check-agent-record.py
  • agent record OK: ENGINE=131 MODEL=327 QUANT=82 KERNEL=45 BACKEND=68
• python3 tests/scripts/test_agent_record.py
  • 13 mutation tests
• python3 scripts/check-doc-checkpoint.py
• python3 tests/scripts/test_doc_checkpoint.py
  • 5 tests
• python3 scripts/check-readme-structure.py
• python3 tests/scripts/test_check_readme_structure.py
  • 18 mutation tests
• git diff --check HEAD^

No source, header, CMake, test, README, model, kernel, fixture, generated artifact, GPU work, model download, or benchmark changed. docs/BENCHMARKS.md records NOT APPLICABLE and benchmark_binding=false.

Slack

The required selection notification was attempted with the bundled secret-safe sender. No channel is configured; the only conventional target available (#general) returned channel_not_found. No credential was read or exposed.

[localai-org-maint-bot]

Slack follow-up: the required PR-open notification was also attempted through the bundled secret-safe sender. With no configured channel, #general again returned channel_not_found; no credential was read or exposed.