229 parse all arguments in pytorch trace file

[fengxizhou]

What does this PR do?

• The HTA currently filters a specific set of arguments defined by the ParserConfig class. This approach effectively minimizes memory footprint, which is crucial when dealing with fixed traces.

• When working with new types of traces, such as MTIA traces, the current implementation requires modifying the ParserConfig class. This introduces an additional layer of effort and potential complexity.

Proposed Solution

We're adding a new boolean attribute, parse_all_arguments (default False), to the ParserConfig class. When set to True, the parser will parse all arguments in the trace file. Arguments defined in events_args will be parsed as specified, while undefined arguments will be parsed using a standard naming convention with inferred default values.

The undefined arguments are parsed with the following method:
name: the converted name, which is the raw name from the trace file converted to lowercase and with spaces replaced by underscores.

• raw_name: the original name as it appeared in the trace file
• value_type, the data type inferred based on the value that was given to the argument
• default_value: -1 for int, "" for string, None for object

Benefits

• Flexibility: Enables seamless adaptation to new trace types without modifying the ParserConfig class.
• Ease of Use: Provides a simple and intuitive way to toggle between parsing specific arguments and parsing all arguments.
• Reduced Maintenance: Simplifies the codebase by eliminating the need for frequent updates to the ParserConfig class for new trace types.

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
  • N/A
• Did you write any new necessary tests?
  • N/A
• Did you make sure to update the docs?
  • N/A
• Did you update the changelog?
  • N/A