Create a training set of labeled repos for DDP

[ccerv1]

We'll iterate on these with DDP, but some initial ideas:

• "abandoned" - project was started and quickly abandoned
• "duplicate" - project is double-counted or a fork that doesn't not deviate from the main project
• "false positive" - project does not seem like it belongs in the dataset
• "spammy" - project has a lot of bot-like activity or other signs of manufactured activity
• "high quality missing from OSO" - high quality projects that are missing from OSSD

[linear[bot]]

OSO-1218