docs(api): OpenAI server supports tool-calling

[jeswr]

docs/api.md (the OpenAI coli serve section) currently says Tools "return an explicit error rather than being silently ignored." That's stale for Tools: c/openai_server.py at HEAD ships full tool-calling — parse_tool_calls() (renders GLM tool markers back into OpenAI tool_calls), render_chat(..., tools, tool_choice) (injects the <tools> XML and honours tool_choice including none/required/forced-function), plus a dedicated test_openai_tools_e2e.py. Tool schemas only 400 when malformed (landed via #401/#408). The Anthropic section of the same doc already documents full tools support, so the OpenAI paragraph contradicted both the code and its sibling section.

This corrects only the Tools sentence; the image/audio input, custom stop sequences, log-probabilities, and token-penalty items are left unchanged.

Prepared for review — happy to adjust wording. (Heads-up: there may be a delay before I can respond to review comments.)

[JustVugg]

Thanks — and this docs fix is correct: tool-calling is supported on the OpenAI /v1/chat/completions path (the previous wording was wrong). I retargeted it from main to dev (all contributions land on dev), which surfaced a conflict in docs/api.md since that file moved on dev. Could you rebase onto current dev and resolve it? Then it merges.

[JustVugg]

dev has moved a fair amount (v1.3.0: Kimi K3 and Inkling engines, the ragged-attention grouped-scale fix, CUDA_RELEASE_HOST default, the Inkling expert-cache fix). This PR now conflicts — could you rebase onto latest dev? Nothing is wrong with the change itself; CI was green where it ran. Ping me once it's clean and I'll review/merge.