Add ConvFinQACalc

[yifanmai]

This is a mathematical calculation benchmark based on ConvFinQA: Exploring the Chain of Numerical Reasoning in Conversational Finance Question Answering (Chen ey al., 2022) from the IBM Enterprise benchmark.

[yifanmai]

Hi @ryokawajp and @mtake, please review this pull request.

By approving, you acknowledge that:

• You will be listed as a co-author of the merge commit
• You agree to license your work under Apache 2.0

Main changes:

• The scenario has been renamed from ConvFinQA to ConvFinQACalc to clarify that the task is different from the original paper's task. The new task is to output a number, rather than to write code for a calculation.
• Scenario implementation has been simplified.
• Scenario description and metadata have been added to the schema.
• The metric has been added to the schema.
• The Githash in the source file URL has been pinned.
• Metrics code has been moved to its own file conv_fin_qa_calc_metrics.py.
• Additional instructions have been added to help instruction following models.
• Prompt formatting has been changed to fix inconsistent whitespace formatting.

[ryokawajp]

Thank you! @yifanmai .
LGTM.