improvements to speech to speech

[adamlam2-amd]

@ldokovic-personal @nikitaandrade-amd

Took feedback from Lazar and myself and modified this playbook. Key changes:

• configured model installed to occur automatically and be managed by HuggingFace so users don't have to do it themselves
• bug fixes in terms of dtype vs torch_dtype so that the code can run without error
• Flow changes; we have the user download all the files first, and then explain each function after that if they wish
• Took out the -share version of gradio as it has complicated unnecessary step (added it to next steps)
• general formatting

[ldokovic-personal]

The difference compared to the PR created `254 which addresses creating a global MODEL_PATH as an env variable is that here we are pulling it out of use across all platforms and OS's. It seems cleaner and more intuitive from a User point of view. In this way, we are making it clear that we will not deal with paths to the user directory in further steps.

Overall Lgtm

[adamlam2-amd]

@sreeram-11 please consider this PR when thinking of tests for speech-speech playbook. Major possible testing changes are:

• downloading the model using Huggingface and its storage artifacts
• took out the gradio_share part