Dump the assessments in the CSV files once they happen and not in the end of all executions

[ruiAzevedo19]

TODO

1st iteration

• [x] Dump the assessments into the evaluation.csv every time a task is executed

2nd iteration

• [x] Create the other CSVs from the evaluation.csv
  • read evaluation.csv > sum the results > write to file
  • [x] models-summed.csv
  • [x] <language>-summed.csv

Follow-ups

• [x] Store the tasks and repositories in the evaluation records
  • https://github.com/symflower/eval-dev-quality/pull/241#discussion_r1666425640
• [ ] Add the model human-readable name and its costs only when the raw records of the evaluation.csv are processed
  • https://github.com/symflower/eval-dev-quality/pull/241#discussion_r1666436342
• [x] Change EvaluationRecordsPerModel to be map[string][]*metrics.Assessments
  • https://github.com/symflower/eval-dev-quality/pull/241#discussion_r1668526526

[bauersimon]

• would basically just have only one evaluation.csv where everything is just dumped on the fly
• the other two csvs can be created from the first one, it should just be different summing
  • thats also already the first step towards "combining multiple evaluations"
  • basically one single "raw" results file, and then we can do all the pretty stuff in a separate step
• basically in this iteration you just read in the raw csv and do the summing up and write them out at the end of the evaluation