Dynamic block device volume size and entry point changes

[estohlmann]

A request was made to support block device volume size being configurable, added optional API param to accomplish this.

For the TGI entrypoint:

• Added number of shards and CUDA settings (increase model memory footprint by distributing model across available GPU's)
• Fixed quantization not being applied by text-generation-launcher (variable being set but not applied by text-generation-launcher)

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.