fix(wasm): wire JS model provider via set_model_provider(fn) — new JsModelProvider(fn) throws after agent-execute-core import

[hobolabsdigital]

Problem

wasm_agent_create / createWasmAgent fail for any model with:

JsModelProvider requires a function argument

Root cause: in attachJsModelProvider, await import('../mcp-tools/agent-execute-core.js') is evaluated, then new mod.JsModelProvider(fn) is constructed — after that module graph loads, the wasm-bindgen ctor rejects its (valid) JS callback. Reproduced in a fresh Node process via createWasmAgent directly (not a stale runtime). @ruvector/rvagent-wasm@0.1.0 constructs fine in isolation, so it's the construct-after-graph-load interaction. Net effect: ADR-129 P1 real-LLM wiring is dead on the current build — every provider-keyed create throws; only the keyless echo stub survives.

Fix

WasmAgent.set_model_provider accepts a raw Function (per rvagent_wasm.d.ts). Skip the broken wrapper and pass the callback directly. One-line-ish change in attachJsModelProvider.

Verification

Fresh Node process, RUFLO_PROVIDER=ollama + OLLAMA_API_KEY, model: "ollama:glm-5.2:cloud": createWasmAgent → promptWasmAgent succeeds; the provider call lands on glm-5.2:cloud at https://ollama.com/v1/chat/completions and returns content (pre-fix: throws).

Secondary note (optional, not in this diff)

callOllamaCompat appends /v1/chat/completions to OLLAMA_BASE_URL, so a base of https://ollama.com/v1 becomes .../v1/v1/.... Docs should specify the bare host (https://ollama.com) for the ruflo provider path. (The env-var table also lists OLLAMA_URL=localhost:11434, but the code reads OLLAMA_BASE_URL defaulting to https://ollama.com.)