Quanitize and Build templates

[sarmiena]

Different models will need different quantizations and trtllm-build flags based on: use case, hardware, expected throughput, etc.

New requirements when running build-engine command:

• allow no json file to be present (which would just use blank values for all quantize/build). this would mean the user needs to supply bare minimum values manually using --quantize- --trtllm-build-. currently we cause an error when this happens
• move json files from script/models/*.json to script/build_templates/$MODEL/<template_name>.json
• Add REPL for asking template preferences

REPL Requirements

• Asks which template you want to use. Only shows available options from based on the model. Choose a number and hit enter
• If you select one from the list, it prints the config and asks for confirmation before running.
• If you say "No", then it reprints options
• Has a option

Stretch goal

Just so the thought isn't lost, it would be nice for users to publish their build/run settings somewhere and explain what model, use case, hardware, versions, benchmarks, for the given config. Be able to pull those down and just run. Maybe even from the REPL