fix(integration tests): Dynamic Ollama model discovery for concurrent multi-model requests

[RyamL1221]

fix: Dynamic Ollama model discovery for concurrent multi-model requests

Problem

When concurrent LLM requests specify different Ollama models (e.g., qwen3:1.7b and qwen3:4b), requests for models not pre-registered in config.yaml fail with a 500 error: "Selected LLMs are not all available" — even though those models are installed and available on the Ollama server.

This caused 2 of 4 concurrent Ollama CI tests to fail (test_ollama_different_models, test_ollama_some_specified_different_models).

Root Cause

LLMAdapter._initialize_llms() only registers models explicitly listed in config.yaml. The availability check (check_availability_for_selected_llm_lists) does a strict name-in-list comparison with no fallback to query the Ollama server for models that are installed but not pre-configured.

Fix

Added dynamic Ollama model discovery to LLMAdapter in aios/llm_core/adapter.py:

• _query_ollama_available_models() — queries the Ollama server's /api/tags endpoint to get installed models
• _dynamic_register_ollama_model() — thread-safe, idempotent method that discovers and registers an Ollama model on-demand (updates llm_configs, llms, available_llm_names, and router config)
• Modified execute_llm_syscalls() — when a request fails the initial availability check and the unavailable models have an "ollama" backend, attempts dynamic registration before rejecting

Non-Ollama backends, already-configured models, and truly unavailable models are completely unaffected.

Files Changed

• aios/llm_core/adapter.py — dynamic registration infrastructure + integration into execute_llm_syscalls()
• aios/config/config.yaml.example — added qwen3:4b to the default Ollama models list (CI pulls both models)
• requirements.txt — added hypothesis (test dependency for property-based tests)
• tests/modules/llm/ollama/test_dynamic_registration_pbt.py — new test file with Hypothesis-based bug condition and preservation tests (runs without Ollama server, fully mocked, compatible with both pytest and unittest runners)

Testing

• 7 property-based tests pass locally (2 bug condition + 5 preservation)
• Tests work with both python -m pytest and python -m unittest (CI uses the latter)
• All 4 concurrent Ollama integration tests pass in CI (test_ollama_same_model, test_ollama_different_models, test_ollama_some_specified_same_model, test_ollama_some_specified_different_models)
• All 4 single Ollama tests continue to pass