[Feature] Generate a software accuracy report

[Udayraj123]

Requirements

Software accuracy report

• We can assume that a ground-truth CSV output is provided, which can be matched with OMRChecker's output and get an accuracy score/report.
• Need to list the erroneous changes and show the marked images for them.
• We can get an accuracy number based on review portal's output (ground truth)

Ideas to start with

• We can match OMRChecker's output with provide ground truth and display an accuracy score
• Create a report from the mismatches to highlight which questions/question blocks were most mismatched(https://github.com/Udayraj123/OMRChecker/issues/39)

Further ideas

This implementation can be extended to create a test benchmark for any existing OMR checking software. We need to blackbox the logic and standardise the inputs and outputs in this case.

[Udayraj123]

Crosslinking discord thread: https://discord.com/channels/590134763784896514/1270815064256938183/1270816309940326572

Types of accuracy metrics
    - csv-based accuracy benchmark (using moderation portal)
    - Needs to fit in the per-directory approach
    - bubble detection/reading level accuracy
        - confidence metrics
    - preProcessor level accuracies
        - confidence metrics for the matchers
    - aggregate metric
    - speed graphs to identify

- Since we have 4-5 confidence metrics and 2-3 accuracy metrics, it'd be great to find a color spectrum that caters to these (color strips of 3 at a time if needed). Where all-white means most accurate and all-black means least accurate, and the weighted magnitude of the color gives the numeric representation.
- Can also make a Radar chart of this for explanation purposes as it's across different dimensions.