evaluate SWT as a image classification task, not time frame annotator

[keighrim]

Because

Although we have a ongoing issue to evaluate the SWT app (#43 ), the proposed evaluation is based on time frame-based scene recognition task, largely because when the issue was brought up, the SWT output only consist of time frame annotations, after binning and stitching.

However, as the SWP app is evolved into separable stitcher and customizable post-binning, the app is now outputs image level "raw" classification results, so we now want to add additional evaluation method for SWT that works at image level.

Done when

• A new name is picked for image classification task subdirectory
• Evaluation metrics for image classification task is determined (most likely simple P/R/F1)
• evaluate.py is written to compute the evaluation metrics given prediction MMIF and gold SR annotation
• The first evaluation is conducted with the latest version of SWT at the time of evaluating, and report is submitted.

Additional context

No response

[keighrim]

fixed mostly by #55, but I don't like the subdirectory name sr_eval, since the name too much bound to a particular image classification task, while the evaluation scheme can be more broadly used for other timepoint-wise P/R/F calculation. Leaving this issue open until #60 is done (to avoid merge conflicts) and a new name is applied to the subdirectory.