Training-noodles

* Fix missing argument error when calling response commands

* Avoid running response commands recursively due to response commands errors
* Log messages first before running response commands

* Changed the default metric output from percentage to a range between 0.0-1.0

* Added requirement `cuda_memory_usage` in the default spec
* Removed requirement `cuda_memory_utilization` in the default spec

* Added spec option `commands_interval` to add a delay between command executions
* Added spec option `error_handlers[].commands` for executing response commands when errors occur
* Added spec options `shell_string` and `shell_stdin` for customizing the shell commands to read from string and STDIN
* Add command type and user spec path in the output file of `write_status_to`

* Transferred the repo to elsa-lab