Make loading models easier

[pweglik]

Right now we need to pass all 3 paths/urls, but they are already premade for most of models configurations, so instead:

 const llama = useLLM({
    modelSource: LLAMA3_2_1B,
    tokenizerSource: LLAMA3_2_TOKENIZER,
    tokenizerConfigSource: LLAMA3_2_TOKENIZER_CONFIG,
  });

we could have:

 const llama = useLLM({
    model: LLAMA3_2_1B,
  });

We should introduce it as a non-breaking change, make all attributes optional and add checks that either model or the previous trio needs to be passed

[chmjkb]

We could just export preset objects, for example:

export const LLAMA_3_2_1B_PRESET =  {
    modelSource: LLAMA3_2_1B,
    tokenizerSource: LLAMA3_2_TOKENIZER,
    tokenizerConfigSource: LLAMA3_2_TOKENIZER_CONFIG,
  }

Then the user would just do:

import { LLAMA_3_2_1B_PRESET } from 'react-native-executorch';

const llm = useLLM(LLAMA_3_2_1B_PRESET)