add ability to instantiate direct client with template

[dineshyv]

needs to go in after https://github.com/meta-llama/llama-stack/pull/477

test plan: 1) pip install -e . on llama stack main 2) pip install -e . in llama stack client python 3) run a app with direct client:

import asyncio
from llama_stack_client.lib.direct.direct import LlamaStackDirectClient

async def main():
    client = await LlamaStackDirectClient.from_template("ollama")

    response = await client.models.list()
    print(response)

if __name__ == "__main__":
    asyncio.run(main())

INFERENCE_MODEL=llama3.2:1b-instruct-fp16 python app.py
Resolved 12 providers
 inner-inference => ollama
 inner-memory => faiss
 models => __routing_table__
 inference => __autorouted__
 inner-safety => llama-guard
 shields => __routing_table__
 safety => __autorouted__
 memory_banks => __routing_table__
 memory => __autorouted__
 agents => meta-reference
 telemetry => meta-reference
 inspect => __builtin__

checking connectivity to Ollama at `http://localhost:11434`...
`llama3.2:1b-instruct-fp16` already registered with `ollama`
Models: llama3.2:1b-instruct-fp16 served by ollama

[Model(identifier='llama3.2:1b-instruct-fp16', provider_resource_id='llama3.2:1b-instruct-fp16', provider_id='ollama', type='model', metadata={})]