Make Architect Eval Benchmark Input Schema Configurable

[humza-sami]

Currently, the Architect class only supports .jsonl benchmark files with a fixed schema containing "question" and "answer" keys. This limits flexibility for users who have datasets with different key names or structures.

We need to add functionality to allow configuration of these keys, so users can specify which key represents the question and which key represents the reference answer when creating an Architect instance.