Add incomplete samples tracking and overall progress to viewer

[sjawhar]

• Currently the viewer has no knowledge of which tasks/samples are still pending, because that is not saved when starting an eval
• So starting an eval should save a file that says "here are all the samples that will be run eventually", which the UI would match against
  • Talk to Charles Teague to align on implementation ahead of time.

[sjawhar]

Inspect does now save an eval-set.json file with list of tasks (not samples) and latest PyPI release includes a progress bar in the viewer