Preparing for exporting answer metadata to euler-frontend

Introduction

We decided that it was best that euler-frontend has access to retrieved chunks. This can help parsing and improve display. In fact, euler-api already exports the metadata (see metadata flag), but at the moment, this is only useful for RAG assessment. In this MR, we make some changes to the api flag behavior so euler-frontend can also exploit this feature.

Manual tests

(see docs update)

No-Reg

$ zcat latest.json.gz | jq '.summary'
{
  "context_precision": {
    "avg_score": 0.8364583333120121,
    "failures": 0,
    "total_queries": 20
  },
  "context_recall": {
    "avg_score": 0.8282142857142858,
    "failures": 0,
    "total_queries": 20
  },
  "correctness": {
    "avg_score": 0.3216392308951872,
    "failures": 2,
    "total_queries": 20
  },
  "faithfulness": {
    "avg_score": 0.8380458722563987,
    "failures": 1,
    "total_queries": 20
  }
}