[Feature] Lambda model runner

[achad4]

While LLMs will not be hosted directly on lambdas, allowing customers to proxy their eval requests through Lambda invocations would expand the breadth of use cases of FMEval, without needing to maintain a vast set of model runner classes. Use cases include:

1. Use cases where customers host LLMs hosted on bare metal (EC2, etc).
2. Use cases where LLMs that are hosted on Sagemaker or Bedrock, but invoked through a web service that performs preprocessing / post-processing.

The Lambda model runner can function identically to the Bedrock model runner, replacing the Bedrock runtime call with a lambda invocation.

[Zhenshan-Jin]

@achad4 thank you for your interest in fmeval. Your request has been shared with the team and will keep you posted about the next step plan.